Structural Engineering
Analysis and Design
_________________________________________________________
SECOND EDITION
TABLE OF CONTENTS
ACKNOWLEDGEMENTS .....................................................................................................................................13
1 INTRODUCTION..............................................................................................................................................17
3.1 GL, ML STATIC ANALYSIS BY THE IMPLICIT DIRECT STIFFNESS METHOD ..............................................42
3.1.1 Mathematical Formulation of Analysis.........................................................................................................................................42
3.1.2 ∆ (KGA From KEA) Analysis ...........................................................................................43
Concepts of Geometric Stiffness or P-∆
3.1.3 ∆ (KG From KE ) Linear Static Analysis – Direct Approach............................................................................48
Performing P-∆ A A
3.1.4 ∆ (KGA From KEA) Linear Static Analysis – Modal Approach; And Hence P-∆
Performing P-∆ ∆ Based Buckling ...................50
3.1.4.1 Objectives ............................................................................................................................................................................................................... 50
3.1.4.2 Mathematical Formulation...................................................................................................................................................................................... 50
3.1.4.3 Critical Load Case Considerations.......................................................................................................................................................................... 52
3.1.4.4 Modelling (Modal) Imperfections........................................................................................................................................................................... 53
3.1.4.5 Design..................................................................................................................................................................................................................... 57
3.1.4.6 Limitations.............................................................................................................................................................................................................. 58
3.1.4.7 Methodology........................................................................................................................................................................................................... 59
3.1.5 MSC.NASTRAN Decks ..................................................................................................................................................................62
3.1.5.1 ML, GL Static Analysis .......................................................................................................................................................................................... 62
3.1.5.2 GL, ML P-∆ (KGA From KEA) Static Analysis......................................................................................................................................................... 71
3.1.5.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Static Analysis ..................................................................................................................... 72
3.1.6 Hand Methods Verification............................................................................................................................................................73
3.1.6.1 Static Displacements by the Unit Load Method of the Virtual Work Principle....................................................................................................... 73
3.1.6.2 BMD and SFD of Structures of Low Static Indeterminacies by Flexibility Analysis.............................................................................................. 76
3.1.6.3 Bending Moments, Shear Force, and Displacements of Structural Elements by the Stiffness Method ................................................................... 98
3.1.6.4 Bending Moments and Shear Force of Structures of Low Kinematic Indeterminacies by Moment Distribution .................................................... 99
3.1.6.5 Summary of Deflections and Effects..................................................................................................................................................................... 123
3.1.6.6 Member Buckling Check Using The P-∆ Method To Incorporate Imperfections, Residual Stresses .................................................................... 164
3.1.6.7 Overall Portal Frame and Multi-Storey Building P-∆ Analysis Using The Amplified Sway Method ................................................................... 177
3.2 GL, ML BUCKLING ANALYSIS BY THE IMPLICIT LINEARIZED EIGENVALUE ANALYSIS .........................178
3.2.1 Linearization of Tangent Stiffness Matrix and Formulating The Linear Eigenvalue Problem .............................................178
2
Structural Engineering Analysis and Design
Second Edition
3.2.2 Problem Reduction and Trial Modes in Linear Buckling Analysis ..........................................................................................184
3.2.3 Concepts of Linearized (KGA From KEA) Buckling Analysis.....................................................................................................185
3.2.4 MSC.NASTRAN Decks ................................................................................................................................................................186
3.2.4.1 GL, ML (KGA From KEA) Buckling Analysis ........................................................................................................................................................ 186
3.2.4.2 GL, ML (KGA From Exact or Approximate KTA) Buckling Analysis .................................................................................................................... 192
3.2.5 Hand Methods Verification..........................................................................................................................................................195
3.2.5.1 Member or Local (KGA From KEA) Buckling ........................................................................................................................................................ 195
3.2.5.2 Overall System (KGA From KEA) Buckling............................................................................................................................................................ 203
3.3 GL, MNL PLASTIC COLLAPSE ANALYSIS BY THE IMPLICIT LINEAR SIMPLEX PROGRAMMING ............214
3.3.1 Mathematical Formulation of Analysis.......................................................................................................................................214
3.3.2 Mathematical Proof ......................................................................................................................................................................217
3.3.3 Displacements and Rotations at Collapse ...................................................................................................................................221
3.3.4 Plastic Limit Analysis of Simple Framed Structures .................................................................................................................222
3.3.4.1 Portal Frame ......................................................................................................................................................................................................... 222
3.3.4.2 Parabolic Arch ...................................................................................................................................................................................................... 228
3.3.4.3 Displacements and Rotations at Collapse.............................................................................................................................................................. 229
3.3.5 Plastic Limit Analysis of Rectangular Multi-Storey Multi-Bay Frames with Improved Data Generation ...........................231
3.3.5.1 Two Storey Frame With Concentrated Loading.................................................................................................................................................... 232
3.3.5.2 Three Storey Frame With Distributed Load .......................................................................................................................................................... 234
3.3.5.3 Multi-Bay Portal Frame ........................................................................................................................................................................................ 237
3.3.5.4 Vierendeel Truss ................................................................................................................................................................................................... 239
3.3.6 Hand Methods Verification..........................................................................................................................................................240
3.3.6.1 Plastic Collapse Analysis by Solving Equilibrium Equations ............................................................................................................................... 240
3.3.6.2 Plastic Collapse Analysis by Virtual Work (Hobbs) ............................................................................................................................................. 244
3.3.6.3 Plastic Collapse Analysis by Virtual Work (Chryssanthopoulos) ......................................................................................................................... 246
3.4 GNL, MNL, CONTACT NONLINEAR STATIC AND BUCKLING ANALYSIS BY THE IMPLICIT TRACING THE
EQUILIBRIUM PATH METHOD ................................................................................................................................262
3.4.1 Mathematical Formulation of Tracing the Equilibrium Path...................................................................................................262
3.4.2 Newton-Raphson Load Control, Displacement Control or Arc-Length Control Algorithm..................................................265
3.4.3 Equilibrium Paths, Stability of Equilibrium Paths, Critical Points, Stability of Critical Points............................................266
3.4.4 MSC.NASTRAN Decks ................................................................................................................................................................269
3.4.4.1 GNL, MNL Load Control, Displacement Control or Arc-Length Control Static Analysis.................................................................................... 269
3.4.4.2 Nonlinear Static and Linearized Eigenvalue Buckling Analysis ........................................................................................................................... 280
3.4.4.3 Nonlinear Static and Linear Eigenvalue Modal Dynamic Analysis ...................................................................................................................... 281
3.4.4.4 Restart From Nonlinear Static Analysis SOL 106 Into Nonlinear Static Analysis SOL 106................................................................................. 282
3.4.4.5 Restart From Nonlinear Static SOL 106 Into Linear Solution Schemes SOL 107 to SOL 112 ............................................................................. 282
3.4.4.6 Implicit Nonlinear Static (and Dynamic Analysis) SOL 400 ................................................................................................................................ 282
3.4.4.7 Implicit Nonlinear Static (and Dynamic Analysis) SOL 600 ................................................................................................................................ 282
3.5 GNL, MNL STATIC AND BUCKLING ANALYSIS BY DYNAMIC RELAXATION.............................................283
3.5.1 Dynamic Relaxation of the Explicit Finite Difference Scheme Solving Newton’s Dynamic Equilibrium ODE (LS-DYNA)
283
3.5.2 LS-DYNA (GNL, MNL Explicit Transient) Dynamic Relaxation Cards.................................................................................286
3
Structural Engineering Analysis and Design
Second Edition
4.2.5.1 Determination of Damped Natural Frequency and the Maximum Dynamic Displacement, umax for Free Damped Vibration Due to Initial
Displacement and/or Initial Velocity by Classically Solving the SDOF Linear ODE and Maximizing the Solution ............................................................. 326
4.3 GL, ML IMPLICIT (REAL) MODAL FREQUENCY RESPONSE ANALYSIS .....................................................329
4.3.1 Nature of the Dynamic Loading Function ..................................................................................................................................329
4.3.2 Mathematical Formulation of Analysis.......................................................................................................................................329
4.3.3 Capability of A Finite Number of Modes To Model The Static and Dynamic Response of Structure...................................334
4.3.4 Complex Response F(ω ω) and Amplification D(ω ω) With Elemental and/or Modal Structural Damping................................337
4.3.5 Representation of A MDOF System As A SDOF system ...........................................................................................................338
4.3.6 MSC.NASTRAN Decks ................................................................................................................................................................340
4.3.6.1 GL, ML Modal Forced Frequency Response Analysis ......................................................................................................................................... 340
4.3.6.2 GL, ML P-∆ (KGA From KEA) Modal Forced Frequency Response Analysis ........................................................................................................ 343
4.3.6.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Modal Forced Frequency Response Analysis .................................................................... 345
4.3.7 Hand Methods Verification..........................................................................................................................................................346
4.3.7.1 Determination of Maximum Dynamic Displacement for Deterministic Frequency Domain Loading by Transforming the Coupled MDOF Linear
Damped ODEs To a Set of Independent (Uncoupled) SDOF ODEs and Solving the Independent Equations in a Manner Similar to Solving a SDOF ODE346
4.4 GL, ML IMPLICIT (COMPLEX) MODAL FREQUENCY RESPONSE ANALYSIS .............................................355
4.4.1 Mathematical Formulation of Analysis.......................................................................................................................................355
4.5 GL, ML IMPLICIT DIRECT FREQUENCY RESPONSE ANALYSIS ..................................................................358
4.5.1 Nature of the Dynamic Loading Function ..................................................................................................................................358
4.5.2 Mathematical Formulation of Analysis.......................................................................................................................................358
4.5.3 MSC.NASTRAN Decks ................................................................................................................................................................361
4.5.3.1 GL, ML Direct Forced Frequency Response Analysis.......................................................................................................................................... 361
4.5.3.2 GL, ML P-∆ (KGA From KEA) Direct Forced Frequency Response Analysis ........................................................................................................ 371
4.5.3.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Direct Forced Frequency Response Analysis..................................................................... 373
4.5.4 Hand Methods Verification..........................................................................................................................................................374
4.5.4.1 The Theory of the Dynamic Magnification Factor for Undamped Motion and Hence the Determination of Maximum Dynamic Displacement, umax
for Deterministic Harmonic Loading by Classically Solving the SDOF Linear Undamped ODE and Maximizing the Solution........................................... 374
4.5.4.2 The Theory of the Dynamic Magnification Factor for Damped Motion and Hence Determination of Maximum Dynamic Displacement, umax for
Deterministic Harmonic Loading by Classically Solving the SDOF Linear Damped ODE and Maximizing the Solution .................................................... 376
4.5.4.3 The Theory of Vibration (Base) Isolation and the Force Transmitted Into Rigid Foundation by the Damped Structure Subjected to Deterministic
Harmonic Loading................................................................................................................................................................................................................. 381
4.5.4.4 The Theory of Vibration (Base) Isolation and the Determination of Maximum Dynamic Displacement, umax for Deterministic Harmonic Support
Motion (Displacement, Velocity or Acceleration) by Classically Solving the SDOF Linear Damped ODE (in Absolute and Relative Terms) and Maximizing
the Solution ........................................................................................................................................................................................................................... 383
4.6 GL, ML FREQUENCY DOMAIN ANALYSIS – DETERMINISTIC AND RANDOM DYNAMIC RESPONSE
ANALYSIS .................................................................................................................................................................386
4.6.1 Mathematical Preliminaries of Representing Dynamic Characteristics in the Frequency Domain ......................................386
4.6.2 GL, ML Vibration Testing ...........................................................................................................................................................387
4.6.2.1 Vibration Testing for Model Correlation .............................................................................................................................................................. 387
4.6.2.2 Vibrating Testing For Analysis Procedure Verification ........................................................................................................................................ 390
4.6.3 GL, ML Steady-State Response of Deterministic Periodic (Not Necessarily Harmonic) Long Duration Excitation Utilizing
Fourier Series (or Generally Utilizing Fast Fourier Transforms FFT) .................................................................................................391
4.6.3.1 Fourier Series........................................................................................................................................................................................................ 391
4.6.3.2 Discrete Fourier Series.......................................................................................................................................................................................... 392
4.6.3.3 Discrete Fourier Series in Complex Notation........................................................................................................................................................ 393
4.6.3.4 Double Sided Discrete Fourier Series in Complex Notation ................................................................................................................................. 394
4.6.3.5 Normalized Double Sided Discrete Fourier Series in Complex Notation.............................................................................................................. 395
4.6.3.6 Symmetrical Normalized Double Sided Discrete Fourier Series in Complex Notation......................................................................................... 396
4.6.3.7 Symmetrical Normalized Single Sided Discrete Fourier Series in Complex Notation .......................................................................................... 397
4.6.3.8 Practicalities of the Specification of the Fast Fourier Transform (FFT) Representation ....................................................................................... 398
4.6.3.9 MSC.NASTRAN Fast Fourier Transform Analysis Methodology........................................................................................................................ 400
4.6.4 GL, ML Steady-State Response of Random, Gaussian, and Stationary (and Ergodic) Excitations Utilizing Power Spectral
Density (PSD) Functions............................................................................................................................................................................403
4.6.4.1 Statistic of Time Domain Function ....................................................................................................................................................................... 403
4.6.4.2 Definition of the Power Spectral Density (PSD)................................................................................................................................................... 407
4.6.4.3 Validity of the PSD Representation ...................................................................................................................................................................... 410
4.6.4.4 Generation and Specification of the PSD.............................................................................................................................................................. 411
4.6.4.5 Statistical Information Provided by the PSD......................................................................................................................................................... 412
4.6.4.6 MSC.NASTRAN Random Analysis Methodology ............................................................................................................................................... 415
4.7 GL, ML IMPLICIT (REAL) MODAL TRANSIENT RESPONSE ANALYSIS ......................................................419
4.7.1 Nature of the Dynamic Loading Function ..................................................................................................................................419
4.7.2 Mathematical Formulation of Analysis.......................................................................................................................................419
4.7.3 Capability of A Finite Number of Modes To Model The Static and Dynamic Response of Structure...................................424
4.7.4 Concepts of Equivalent Static Force ...........................................................................................................................................433
4.7.5 Structural Damping in Time Domain Analyses..........................................................................................................................435
4.7.6 MSC.NASTRAN Decks ................................................................................................................................................................437
4.7.6.1 GL, ML Modal Forced Transient Response Analysis ........................................................................................................................................... 437
4.7.6.2 GL, ML P-∆ (KGA From KEA) Modal Forced Transient Response Analysis.......................................................................................................... 440
4
Structural Engineering Analysis and Design
Second Edition
4.7.6.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Modal Forced Transient Response Analysis ...................................................................... 442
4.7.7 Hand Methods Verification..........................................................................................................................................................443
4.7.7.1 Determination of Maximum Dynamic Displacement for Deterministic Time Domain Loading by Transforming the Coupled MDOF Linear
Undamped ODEs To a Set of Independent (Uncoupled) SDOF ODEs and Solving the Independent Equations in a Manner Similar to Solving a SDOF ODE
443
4.7.7.2 Determination of Max Dynamic Displacement for Deterministic Time Domain Support Motion (Displacement, Velocity or Acceleration) by
Transforming the Coupled MDOF Linear Undamped ODEs (In Relative Terms) To a Set of Independent (Uncoupled) SDOF ODEs (In Relative Terms) and
Solving the Independent Equations in a Manner Similar to Solving a SDOF ODE ............................................................................................................... 452
4.7.7.3 Determination of Maximum Dynamic Displacement for Deterministic Time Domain Loading by Transforming the Coupled Distributed System
Linear Damped ODEs (from the governing PDE) To A Set of Independent (Uncoupled) SDOF ODEs and Solving the Independent Equations in a Manner
Similar to Solving a SDOF ODE ........................................................................................................................................................................................... 456
4.8 GL, ML IMPLICIT DIRECT TRANSIENT RESPONSE ANALYSIS ...................................................................457
4.8.1 Nature of the Dynamic Loading Function ..................................................................................................................................457
4.8.2 Mathematical Formulation of Analysis.......................................................................................................................................457
4.8.3 MSC.NASTRAN Decks ................................................................................................................................................................460
4.8.3.1 GL, ML Direct Forced Transient Response Analysis............................................................................................................................................ 460
4.8.3.2 GL, ML P-∆ (KGA From KEA) Direct Forced Transient Response Analysis .......................................................................................................... 470
4.8.3.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Direct Forced Transient Response Analysis ...................................................................... 472
4.8.4 Hand Methods Verification..........................................................................................................................................................473
4.8.4.1 Determination of Maximum Dynamic Displacement, umax by Solving the SDOF Undamped/Damped Linear Equation of Motion ODE for
Deterministic Time Domain Loading With/Without Initial Conditions Using the Convolution Integral (Duhamel’s integral) ............................................. 473
4.8.4.2 Determination of Maximum Dynamic Displacement, umax by Solving the SDOF Undamped/Damped Linear Equation of Motion ODE (In
Relative Terms) for Deterministic Time Domain Support Motion (Displacement, Velocity or Acceleration) With/Without Initial Conditions Using the
Convolution Integral (Duhamel’s integral) ............................................................................................................................................................................ 480
4.9 GNL, MNL IMPLICIT AND EXPLICIT DIRECT TRANSIENT RESPONSE ANALYSIS .....................................481
4.9.1 Nature of the Dynamic Loading Function ..................................................................................................................................481
4.9.2 Deterministic Non-Periodic Short Duration Impulse (a.k.a. Blast) Loading Functions With Subsequent Wave Propagation
482
4.9.3 Projectile Crash (a.k.a. Impact) (and Impulsive Blast) Analysis With Subsequent Wave Propagation................................483
4.9.4 Brittle Snap or Redundancy Check Excitation ..........................................................................................................................486
4.9.5 GNL, MNL Explicit Central FD Scheme for Newton’s Dynamic Equilibrium ODE (LS-DYNA).........................................491
4.9.5.1 Solution of Partial or Ordinary Differential Equations Using Finite Difference (FD) Schemes ............................................................................ 491
4.9.5.2 Mathematical Formulation of Analysis – Explicit Central Finite Difference Scheme........................................................................................... 492
4.9.5.3 Stability................................................................................................................................................................................................................. 493
4.9.5.4 Accuracy............................................................................................................................................................................................................... 494
4.9.6 GNL, MNL Implicit Newmark Scheme for Newton’s Dynamic Equilibrium ODE (MSC.NASTRAN)................................495
4.9.6.1 Mathematical Formulation of Analysis – Implicit Newmark Scheme................................................................................................................... 495
4.9.6.2 Stability................................................................................................................................................................................................................. 497
4.9.6.3 Accuracy............................................................................................................................................................................................................... 497
4.9.7 Comparison Between Implicit and Explicit Time Integration Schemes...................................................................................498
4.9.8 MSC.NASTRAN Decks ................................................................................................................................................................500
4.9.8.1 GNL, MNL Direct Forced (Implicit) Transient Response Analysis ...................................................................................................................... 500
4.9.8.2 Nonlinear Static Analysis and Nonlinear Transient Analysis................................................................................................................................ 504
4.9.8.3 Restart From Nonlinear Static Analysis SOL 106 Into Nonlinear Transient Analysis SOL 129 ........................................................................... 504
4.9.8.4 Restart From Nonlinear Transient Analysis SOL 129 Into Nonlinear Transient Analysis SOL 129 ..................................................................... 504
4.9.8.5 Implicit Nonlinear (Static and) Dynamic Analysis SOL 400 ................................................................................................................................ 505
4.9.8.6 Implicit Nonlinear (Static and) Dynamic Analysis SOL 600 ................................................................................................................................ 505
4.9.8.7 Explicit Nonlinear Dynamic Analysis SOL 700 ................................................................................................................................................... 505
4.9.9 LS-DYNA Decks ...........................................................................................................................................................................506
4.9.9.1 GNL, MNL Direct Forced (Explicit) Transient Response Analysis ...................................................................................................................... 506
4.9.10 Hand Methods Verification........................................................................................................................................................508
4.9.10.1 Determination of Displacement Response Time History by Solving the SDOF Nonlinear (in Stiffness, Damping and Displacement) Equation of
Motion ODE for Deterministic Time Domain Loading With/Without Initial Conditions by Implicit Newmark-β Time Integration Schemes ..................... 508
5
Structural Engineering Analysis and Design
Second Edition
6
Structural Engineering Analysis and Design
Second Edition
5.2.5 Crowd Induced (Synchronous) Vertical Footfall Vibrations for Gymnasium Floors, Dance Floors, Stadium Tiers, Theatre
Tiers and Balconies ...................................................................................................................................................................................630
5.2.6 Crowd Induced (Synchronous) Lateral Vibrations ..................................................................................................................633
5.3 GL, ML DYNAMIC INFLUENCE LINE ANALYSIS FOR TRAIN EXCITATIONS ON BRIDGES.........................636
5.4 GL, ML FLUID (WIND & WATER) - STRUCTURE INTERACTION (AEROELASTIC RESPONSE ANALYSIS) 639
5.4.1 Elementary Hydraulics.................................................................................................................................................................640
5.4.1.1 Pipeline Design (Application of the Steady and Unsteady Bernoulli’s Equation)................................................................................................. 640
5.4.1.2 Elementary Wave Mechanics (Small Amplitude Wave Theory)........................................................................................................................... 725
5.4.2 Static Aeroelastic Response..........................................................................................................................................................732
5.4.2.1 Along-Flow Direction Drag (Pressure Drag and Skin-Friction Drag) (Limited Amplitude Response) ................................................................. 733
5.4.2.2 Across-Flow Direction Lift (Pressure Lift and Skin-Friction Lift) (Limited Amplitude Response) ...................................................................... 735
5.4.2.3 Non-Oscillatory Torsional Divergence (Divergent Amplitude Response) ............................................................................................................ 736
5.4.3 Dynamic Aeroelastic Response ....................................................................................................................................................740
5.4.3.1 Along- and Across-Flow Direction Unsteady Inertial Forces on General Submerged Structures Due to Water Waves (Limited Amplitude
Response) 740
5.4.3.2 Along- and Across-Flow Direction Unsteady Inertial Forces on Cylindrical Submerged Structures Due to Water Waves (And Along-Flow
Direction Drag Forces Due to Steady Currents) (Limited Amplitude Response)................................................................................................................... 741
5.4.3.3 Along-Flow Direction Gust Response (And Along-Flow Direction Drag Forces Due to Steady Mean Wind) (Limited Amplitude Response) ... 744
5.4.3.4 Along-Flow Direction Buffeting Response (Limited Amplitude Response) ......................................................................................................... 761
5.4.3.5 Across-Flow Direction Von Karman Vortex Shedding Induced Response (Limited Amplitude Response) ......................................................... 762
5.4.4 Dynamic Aeroelastic Stability......................................................................................................................................................777
5.4.4.1 Across-Flow Direction Galloping and Stall Flutter (Divergent Amplitude Response).......................................................................................... 777
5.4.4.2 Across-Flow Direction Classical Flutter (Divergent Amplitude Response) .......................................................................................................... 788
5.5 GL, ML VIBRATION FATIGUE ANALYSIS.....................................................................................................796
5.5.1 Maximum Absolute Principal Stress or Strain Field Generation in the Time or Frequency Domain...................................797
5.5.1.1 Choice of Analysis Method................................................................................................................................................................................... 797
5.5.1.2 GL, ML Pseudo-Static Analysis ........................................................................................................................................................................... 798
5.5.1.3 GL, ML Transient Analysis .................................................................................................................................................................................. 799
5.5.1.4 GL, ML Modal Superposition Transient Analysis ................................................................................................................................................ 800
5.5.1.5 GL, ML Frequency Domain Stationary (and Ergodic) Random Analysis............................................................................................................. 801
5.5.2 Stress-Life (S-N), Strain-Life (E-N) or Crack Propagation LEFM Uniaxial Fatigue Analysis ..............................................802
5.5.2.1 Stress-Life (S-N) (Total Life) Uniaxial Fatigue Analysis ..................................................................................................................................... 803
5.5.2.2 Strain-Life (E-N) (Crack Initiation) Uniaxial Fatigue Analysis ............................................................................................................................ 812
5.5.2.3 Crack Propagation LEFM Uniaxial Fatigue Analysis ........................................................................................................................................... 813
5.5.3 Stress-Life (S-N), Strain-Life (E-N) or Crack Propagation LEFM Multi-axial Fatigue Analysis .........................................814
5.6 GL, ML PASSIVE STRUCTURAL MOTION CONTROL ...................................................................................815
5.6.1 Optimum Stiffness and Mass Distribution..................................................................................................................................815
5.6.1.1 Concepts of Forced Frequency Response of Deterministic Periodic Harmonic Load Excitations......................................................................... 815
5.6.1.2 Concepts of Forced Transient Response of Deterministic Load Excitations ......................................................................................................... 818
5.6.1.3 Concepts of Forced Frequency Response of Deterministic Periodic Harmonic Base Excitations ......................................................................... 822
5.6.1.4 Concepts of Forced Transient Response of Deterministic Base Excitations.......................................................................................................... 823
5.6.1.5 Concepts of Forced Transient Response (Response Spectrum Analysis) of Random Non-Stationary Base Excitations ....................................... 824
5.6.2 Optimum Damping Distribution .................................................................................................................................................825
5.6.2.1 Elemental Damping Mathematical Models ........................................................................................................................................................... 825
5.6.2.2 Modal Damping Mathematical Models................................................................................................................................................................. 834
5.6.2.3 Global Damping Mathematical Models ................................................................................................................................................................ 837
5.6.2.4 Damping Formulation Conclusions ...................................................................................................................................................................... 839
5.6.3 GL, ML Base Isolation Systems...................................................................................................................................................840
5.6.3.1 Controlling Displacement Response From Harmonic Load Excitations ............................................................................................................... 840
5.6.3.2 Controlling Acceleration Response From Harmonic Load Excitations................................................................................................................. 840
5.6.3.3 Base Isolation - Controlling Displacement Response From Harmonic Base Enforced Motion ............................................................................. 841
5.6.3.4 Base Isolation - Controlling Acceleration Response From Harmonic Base Enforced Motion .............................................................................. 842
5.6.3.5 Base Isolation – Controlling Force Transmitted Into Foundation From Harmonic Load Excitations.................................................................... 844
5.6.4 GL, ML Tuned Mass Damper (TMD) Systems ..........................................................................................................................845
5.6.4.1 Damped SDOF System Subject to Harmonic Force and Support Motion Excitations .......................................................................................... 848
5.6.4.2 Undamped Structure, Undamped TMD System Subject to Harmonic Force Excitation ....................................................................................... 849
5.6.4.3 Undamped Structure, Damped TMD System Subject to Harmonic Force Excitation ........................................................................................... 851
5.6.4.4 Undamped Structure, Damped TMD System Subject to Harmonic Support Motion ............................................................................................ 857
5.6.4.5 Damped Structure, Damped TMD System Subject to Harmonic Force and Support Motion................................................................................ 859
5.6.4.6 TMD Analysis Summary (Warburton, 1982)........................................................................................................................................................ 861
5.6.5 GL, ML Tuned Slosh Damper (TSD) Systems ..........................................................................................................................863
5.6.5.1 When To Employ TSDs........................................................................................................................................................................................ 863
5.6.5.2 Operations of TSDs .............................................................................................................................................................................................. 863
5.6.5.3 Deep or Shallow Tank?......................................................................................................................................................................................... 871
5.6.5.4 Testing .................................................................................................................................................................................................................. 872
5.6.5.5 Miscellaneous Design Information ....................................................................................................................................................................... 873
5.6.5.6 Design Procedure for Deep Tank TSDs ................................................................................................................................................................ 874
5.7 GL, ML ACTIVE STRUCTURAL MOTION CONTROL – CONTROL SYSTEM ANALYSIS...............................875
5.8 FIBRE REINFORCED POLYMER (FRP) ANALYSIS AND DESIGN ..................................................................876
7
Structural Engineering Analysis and Design
Second Edition
8
Structural Engineering Analysis and Design
Second Edition
9
Structural Engineering Analysis and Design
Second Edition
6.9.3 Linear, Elastic, Orthotropic Elastic Material for Shell Elements MAT8 and Solid Elements MAT9 .................................1031
6.10 RIGID ELEMENT CARDS ............................................................................................................................1032
6.11 BOUNDARY CONDITIONS ...........................................................................................................................1035
6.11.1 Single Point Constraints (SPCs) ..............................................................................................................................................1035
6.12 LINEAR OPTIMIZATION SOL 200 .............................................................................................................1036
6.12.1 Objective Function and Constraints........................................................................................................................................1036
6.12.2 Design Variables and Constraints ...........................................................................................................................................1036
6.12.3 Optimization Control Parameters ...........................................................................................................................................1037
6.13 COMPUTATIONAL MEMORY AND PROCESSING POWER DEMAND..........................................................1037
10
Structural Engineering Analysis and Design
Second Edition
8.1.2.1 Universal Beam – Universal Column (No Pinned or Two-Pinned) Portal Frame ............................................................................................... 1059
8.1.2.2 Universal Beam – Universal Column (Three-Pinned) Portal Frame ................................................................................................................... 1062
8.1.2.3 Truss – Stanchion Portal Frame .......................................................................................................................................................................... 1063
8.1.3 Steel and Reinforced Concrete Multi-Storey (Potentially Tall) Buildings .............................................................................1067
8.1.3.1 Building Geometry and Anatomy ....................................................................................................................................................................... 1067
8.1.3.2 Vertical Gravity and Lateral Stability Systems ................................................................................................................................................... 1070
8.1.3.3 In-Plane Stability (Diaphragm Action) ............................................................................................................................................................... 1082
8.1.3.4 Torsional Stability System .................................................................................................................................................................................. 1085
8.1.3.5 Movement Joints................................................................................................................................................................................................. 1086
8.1.3.6 Transfer Structures.............................................................................................................................................................................................. 1087
8.1.3.7 Structural Floor Systems..................................................................................................................................................................................... 1099
8.1.4 Steel, Reinforced Concrete and Masonry Bridges....................................................................................................................1100
8.1.4.1 Steel and Reinforced Concrete Prismatic (Box or Plate) Girder Bridge.............................................................................................................. 1101
8.1.4.2 Steel and Reinforced Concrete Haunched (Box or Plate) Girder Bridge............................................................................................................. 1102
8.1.4.3 Steel Truss Bridge............................................................................................................................................................................................... 1103
8.1.4.4 Steel Portal Bridge .............................................................................................................................................................................................. 1107
8.1.4.5 Steel, Reinforced Concrete or Masonry Arch Bridge.......................................................................................................................................... 1108
8.1.4.6 Steel Cable Stayed Bridge................................................................................................................................................................................... 1109
8.1.4.7 Steel Suspension Bridge ..................................................................................................................................................................................... 1110
8.1.5 Steel Towers ................................................................................................................................................................................1112
8.1.6 Steel Masts...................................................................................................................................................................................1113
8.1.7 Steel Space Roofs ........................................................................................................................................................................1114
8.1.8 Steel Cable Roofs ........................................................................................................................................................................1115
8.1.9 Steel Grandstands .......................................................................................................................................................................1116
8.1.9.1 Cantilever Grandstand ........................................................................................................................................................................................ 1116
8.1.9.2 Front Girder Grandstand ..................................................................................................................................................................................... 1116
8.1.9.3 Cable Suspended Grandstand.............................................................................................................................................................................. 1116
8.1.10 Reinforced Concrete Shell Roofs .............................................................................................................................................1117
8.1.11 Timber Single-Storey Buildings...............................................................................................................................................1118
8.1.11.1 Beam – Column (No Pinned or Two-Pinned) Portal Frame.............................................................................................................................. 1118
8.1.11.2 Beam – Column (Three-Pinned) Portal Frame.................................................................................................................................................. 1119
8.1.11.3 Truss – Stanchion Portal Frame ........................................................................................................................................................................ 1120
8.1.12 Timber Multi-Storey Buildings................................................................................................................................................1139
8.1.12.1 Timber Floors ................................................................................................................................................................................................... 1139
8.1.13 Masonry Single- and Multi-Storey Buildings .........................................................................................................................1140
8.1.13.1 Lateral Stability Systems .................................................................................................................................................................................. 1140
8.2 APPRECIATION OF LOADING, ULS AND SLS LOAD CASES .......................................................................1141
8.2.1Load Cases...................................................................................................................................................................................1141
8.2.2Load Factors ...............................................................................................................................................................................1147
8.2.3Load Combinations ....................................................................................................................................................................1148
8.3 STANDARD CONSTRUCTION MATERIAL PROPERTIES ...............................................................................1150
8.3.1 Steel, Aluminium and Concrete.................................................................................................................................................1150
8.3.2 Stainless Steel ..............................................................................................................................................................................1152
8.3.3 Cast Iron......................................................................................................................................................................................1156
8.3.4 Wrought Iron ..............................................................................................................................................................................1157
8.4 REINFORCED CONCRETE DESIGN TO BS 8110-PART 1:1997 (BUILDINGS) AND BS 5400-PART 4 (BRIDGES)
1159
8.4.1 ULS Rectangular Beam Design – Bending, Shear and Torsion Effects .................................................................................1159
8.4.1.1 Bending Effects .................................................................................................................................................................................................. 1159
8.4.1.2 Shear and Torsion Effects ................................................................................................................................................................................... 1164
8.4.2 Concepts of Moment Redistribution .........................................................................................................................................1167
8.4.3 ULS Structural Floor Design – Reinforced Concrete ..............................................................................................................1176
8.4.4 ULS Structural Floor Design – Prestressed Concrete..............................................................................................................1177
8.4.4.1 (Precast) Pre-Tensioned ...................................................................................................................................................................................... 1177
8.4.4.2 (Cast Insitu) Post-Tensioned Concrete Floor Systems ........................................................................................................................................ 1177
8.4.5 ULS Column Design – Bending, Axial, Shear and Torsion Effects ........................................................................................1178
8.4.6 ULS Wall Design.........................................................................................................................................................................1179
8.4.7 ULS Connection Design..............................................................................................................................................................1180
8.4.8 ULS Fire Protection Design .......................................................................................................................................................1181
8.4.9 SLS Durability Design ................................................................................................................................................................1182
8.4.10 SLS Deflection Computation and Criteria..............................................................................................................................1183
8.4.11 SLS Crack Width Calculations................................................................................................................................................1193
8.5 PRESTRESSED CONCRETE ANALYSIS AND DESIGN ....................................................................................1194
8.6 STEEL DESIGN TO BS 5950-PART 1:2000 (BUILDINGS) .............................................................................1244
8.6.1 ULS Beam and Column Design – Bending (Incorporating High Shear Effects), Axial, Shear and Buckling Effects ........1244
8.6.2 ULS Plate Girder Beam and Column Design – Bending (Incorporating High Shear Effects), Axial, Shear and Buckling
Effects 1246
11
Structural Engineering Analysis and Design
Second Edition
8.6.3 ULS Structural Floor Design – Composite Steel and Reinforced Concrete...........................................................................1247
8.6.3.1 Floor Framing ..................................................................................................................................................................................................... 1247
8.6.3.2 Simple Steel I-Beam With Non-Composite or Composite Precast (Hollow) Slabs ............................................................................................. 1248
8.6.3.3 Prismatic Steel I-Beam Encased in T-section Reinforced Concrete, Flat Part of the T- Above the Top Flange of I-Beam................................. 1248
8.6.3.4 Prismatic Steel I-Beam Shear Studded Into Flat-Soffit Reinforced Concrete Slab.............................................................................................. 1248
8.6.3.5 Prismatic Steel I-Beam Shear Studded Into (Parallel Or Perpendicular Ribbed) Metal Deck With Reinforced Concrete................................... 1249
8.6.3.6 Prismatic Steel Plate Girder Shear Studded into Flat-Soffit Reinforced Concrete or (Parallel or Perpendicular Ribbed) Metal Deck with
Reinforced Concrete............................................................................................................................................................................................................ 1250
8.6.3.7 Prismatic Steel I-Beam With Stub Girder And Reinforced Concrete Slab .......................................................................................................... 1250
8.6.3.8 Prismatic Steel I-Beam With Web Openings, Castellated Beam, Cellular Beam or Westok Beam Shear Studded into Flat-Soffit Reinforced
Concrete or (Parallel or Perpendicular Ribbed) Metal Deck with Reinforced Concrete ...................................................................................................... 1251
8.6.3.9 Prismatic Steel Warren (or Modified Warren) Truss or Pratt Truss, Both With or Without a Vierendeel Panel in the Middle (for the Main
Mechanical Duct) Shear Studded into Flat-Soffit Reinforced Concrete or (Parallel or Perpendicular Ribbed) Metal Deck with Reinforced Concrete....... 1258
8.6.3.10 Nonprismatic Tapered (Shallower at the Supports) Steel Plate Girder Shear Studded Into (Parallel or Perpendicular Ribbed) Metal Deck with
Reinforced Concrete............................................................................................................................................................................................................ 1258
8.6.4
ULS Composite Column Design ................................................................................................................................................1259
8.6.5
ULS Connection Design..............................................................................................................................................................1260
8.6.6
ULS Bearing and Joints Design .................................................................................................................................................1261
8.6.7
ULS Brittle Fracture Design ......................................................................................................................................................1262
8.6.8
ULS Fatigue Design ....................................................................................................................................................................1262
8.6.9
ULS Fire Protection Design .......................................................................................................................................................1263
8.6.10
SLS Human Induced Vibration ...............................................................................................................................................1265
8.6.11
SLS Deflection Criteria ............................................................................................................................................................1266
8.6.12
SLS Corrosion Protection ........................................................................................................................................................1268
8.7 STEEL DESIGN TO BS 5400-PART 3 (BRIDGES)..........................................................................................1270
8.8 TIMBER DESIGN TO EC5 .............................................................................................................................1271
8.8.1 ULS Beam and Column Design .................................................................................................................................................1273
8.8.2 ULS Dowel (Nailed, Screwed, Bolted, Doweled) Connection Design......................................................................................1274
8.8.3 ULS Fire Protection Design .......................................................................................................................................................1275
8.9 MASONRY DESIGN TO BS 5628 ...................................................................................................................1276
9 GEOTECHNICAL ENGINEERING...........................................................................................................1277
10 CONVERSION FACTORS.........................................................................................................................1348
BIBLIOGRAPHY.................................................................................................................................................1349
12
Structural Engineering Analysis and Design
Second Edition
ACKNOWLEDGEMENTS
My humble gratitude to the Almighty, to Whom this and all work is dedicated.
A special thank you also to my teachers at Imperial College of Science, Technology and Medicine, London and my
fellow engineering colleagues at Ove Arup and Partners London and Ramboll Whitbybird London.
13
Structural Engineering Analysis and Design
Second Edition
Elemental Notations
m = SDOF mass
[M] = Global MDOF mass matrix
c = SDOF viscous damping constant
[C] = Global MDOF viscous damping matrix
k = SDOF stiffness
[K] = Global MDOF stiffness matrix
u = SDOF displacement
{u},{U}= Global MDOF displacement matrix
{P} = Global nodal loading vector
Mi, [M] = Modal (generalized) mass and modal (generalized) mass matrix
Ci, [C] = Modal (generalized) damping and modal (generalized) damping matrix
Ki, [K] = Modal (generalized) stiffness and modal (generalized) stiffness matrix
ξi, {ξi} = Modal displacement response and modal displacement response vector
SDOF Time Domain Loading and Transient and Steady-State Response Notations
14
Structural Engineering Analysis and Design
Second Edition
P(t) = SDOF Time domain harmonic loading function, P(t) = Real [ P(ω)eiωt ]
P(ω) = SDOF frequency domain complex harmonic loading function
p0 = SDOF harmonic loading amplitude
p0/k = SDOF static displacement
D(ω) = SDOF (magnitude of the) dynamic amplification factor
Dresonant = SDOF (magnitude of the) dynamic amplification factor at resonance when ω = ωn
Dmax = SDOF maximum (magnitude of the) dynamic amplification factor when ω = ωn(1-2ζ2)1/2
F(ω) = SDOF complex displacement response function (FRF), F(ω) = D(ω)(p0/k)e−iθ
H(ω) = SDOF transfer function, H(ω) = D(ω)(1/k)e−iθ
Fresonant = SDOF complex displacement response function at resonance, Fresonant = Dresonant(p0/k)e−iθ
Fmax = SDOF complex maximum displacement response function, Fmax = Dmax(p0/k)e−iθ
u(t) = SDOF time domain displacement response, u(t) = Real [ F(ω)eiωt ]
uresonant = SDOF time domain displacement response at resonance, u(t) = Real [ Fresonant eiωt]
umax = SDOF time domain maximum displacement response, u(t) = Real [ Fmax eiωt]
Tr = SDOF transmissibility of displacement, acceleration or force
{P(t)} = Time domain harmonic loading function vector, {P(t)} = Real [ {P(ω)} eiωt ]
{P(ω)} = Frequency domain complex harmonic loading function vector
Pi(ω) = Modal frequency domain complex harmonic loading function vector, Pi(ω) = {φi}T {P(ω)}
p0i = Modal harmonic loading amplitude
p0i/Ki = Modal static displacement
Di(ω) = Modal (magnitude of the) dynamic amplification factor
Di resonant = Modal (magnitude of the) dynamic amplification factor at resonance when ω = ωni
Di max = Modal maximum (magnitude of the) dynamic amplification factor when ω = ωni(1-2ζi2)1/2
ξi(ω) = Modal complex displacement response function (FRF), ξi(ω) = Di(ω)p0i/Ki e− iθi
ξi resonant = Modal complex displacement response function at resonance, ξi resonant = Di resonant p0i/Ki e− iθi
ξi max = Modal complex maximum displacement response function, ξi max = Di max p0i/Ki e−iθi
{u(t)} = Time domain displacement response vector, {u(t)} = Real [ [Φ]{ξ(ω)}eiωt ]
Additional Abbreviations
15
Structural Engineering Analysis and Design
Second Edition
16
Structural Engineering Analysis and Design
Second Edition
1 INTRODUCTION
This paper serves as a guide to performing effective static and dynamic structural analyses with MSC.NASTRAN
in particular, although the concepts described herein are also equally applicable to other analysis codes.
A disbeliever wonders if the answer provided by a finite element analysis is correct. The question is not whether
the correct answer has been obtained, but rather if the correct question was asked, for the answer will always be
correct.
17
Structural Engineering Analysis and Design
Second Edition
The elemental stiffness formulation of the finite element describes the stiffness behavior of an element. The finite
element elemental stiffness formulation describes the force-displacement relationship of a finite number of degrees
of freedom of the element. This relationship can be formulated by one of two general methods as described below.
2.1.1.1 The Principle of Virtual Displacements of The Principle of Virtual Work (Equivalent to the
Variational Method or The Principle of Minimum Potential Energy)
The stiffness of a linear system is based on the stiffness at the initial undeformed state. Define state A as the initial
undeformed state. The variation of displacement within a finite element can be presented as a function of the matrix
of shape functions [N] and the discrete nodal degree of freedom displacement vector {d} as
{y} = [N]{d}
Vector {y} describes the general displacement function (interpolation function) within the finite element, {d} is the
unknown nodal DOFs and [N] are their corresponding shape functions.
The general strain vector {ε} in terms of {d} can then be derived as the strain is some derivative function of the
displacement function and hence the nodal DOFs {d}
{ε} = [BA]{d}
The general stress vector {σ} can then be established in terms of the strains and hence be expressed in terms of the
DOFs {d} amongst other terms
The second term is due to the initial stresses (such as residual stresses within the element) and the third term is the
initial strains due to temperature shrinkage or lack of fit.
Two equivalent fundamental theorems of structural analysis are the principle of virtual displacements
(virtual work) and the principle of minimum potential energy. These theorems provide the fundamental
Newton’s equilibrium equations for the finite element. The principle of virtual displacements (virtual work) states
that a structural system is in equilibrium in its deflected configuration if the external work performed by the applied
loading over any possible infinitesimal displacement mode is equal to the internal work performed by the
component forces over the corresponding compatible infinitesimal deformations. In other words, a system is in
equilibrium when the external work done equates the internal work. The equivalent principle of minimum potential
energy states that a structural system is in equilibrium in its deflected configuration if its total potential energy (V),
consisting of the system strain energy (U) and the loading potential energy (-W), is stationary with respect to any
infinitesimal variation in the possible deformation modes.
For a finite element, the equilibrium equation is thus derived by equating the external work to the internal energy
for a virtual displacement set.
(Virtual Nodal Disp) x (Real Nodal Forces) + (Virtual General Disp) x (Real Distributed Forces)
= (Virtual Strains Or Deformations) x (Real Stresses Or Actions)
18
Structural Engineering Analysis and Design
Second Edition
where {f} is the nodal force vector and {b} is the distributed body forces within the finite element. We have
however established that
The above expression is valid for any virtual displacement δ{d}. Choosing unity virtual displacements we arrive at
the elemental equilibrium equations
[
{f } = ∫Ω [B A ]T [D A ][B A ]dΩ {d} +] {∫ [B Ω
] {σ}i dΩ −
A T
} {∫ [B
Ω
A T
] [D A ]{ε}i dΩ −} {∫ [N] {b}dΩ}
Ω
Τ
This is the expression that is set up for each and every element. If the nodal force vector {f} is known, then the only
remaining unknown within these equations are the nodal displacement vector {d}. Note that we have defined [B]
the strain matrix, {σ}i = the initial stresses, [D] the material constitutive matrix, {ε}i = the initial strains, [N] =
shape functions matrix, {b} = the elemental (external) body loads and of course {d} = nodal element displacement
in element axes.
To summarize the terms in the above elemental static equilibrium equation expression
(i) the term on the LHS is the nodal force vector in element axis
(ii) the first term on the RHS is the instantaneous stiffness matrix
(iii) the second, third and fourth terms on the RHS are the so-called fixed end forces
We note that the choice of the shape functions [N] affects the accuracy of
The important concept to grasp is that geometrically linear (GL) finite elements have linear strain-displacement
relationships i.e. the [BA] matrix would be constant and thus independent of the nodal displacements {d}.
Materially linear (ML) finite elements have linear stress-strain relationships i.e. the [DA] matrix would be constant
and thus independent of the strain vector {ε}.
19
Structural Engineering Analysis and Design
Second Edition
We have seen that the principle of virtual displacements or the principle of minimum potential energy (variational
approach) can be used to establish the element stiffness matrix. The shape function [N] is applied to the variational
statement in the discretization.
Another method of establishing the elemental stiffness matrix is the method of weighted residuals. Here, instead of
applying the shape functions [N] to the variational statement, it is applied to the so-called weak statement. The
weak statement is a perfectly equivalent integral form of the strong statement which is the governing Newton’s
differential equation representing the equilibrium of the element internal and external forces. The weak statement of
a differential equation (strong statement) can be stated as
Two common methods within the method of weighted residuals are the sub-domain collocation method and the
popular Galerkin method (which is equivalent to the variational approach). The sub-domain collocation method
divides the geometric domain into as many sub-domains as there are DOFs with the weighting functions having a
value of unity in a particular sub-domain and zero in all other sub-domains leading to one equation from the weak
statement. The weighting functions are also put to unity at other sub-domains in turn, leading to a set of
simultaneous equations which formulate the elemental stiffness matrix.
In the Galerkin approach, the weighting functions are identical to the shape functions. Since there are as many
shape functions as there as DOFs, each shape function is applied to the weak statement for an equation. This leads
to a set of simultaneous equations which formulate the elemental stiffness matrix. The Galerkin approach is
actually exactly equivalent to the variational approach.
20
Structural Engineering Analysis and Design
Second Edition
The following describes the assembly of the element stiffness matrices towards the formulation of the global
system stiffness matrix by the use of the equilibrium, compatibility and constitutive laws. We have shown that the
elemental equilibrium equations are
[ ]
= k AE {d} + {∫ [B
Ω
] {σ} dΩ}− {∫ [B ] [ D ]{ε} dΩ}− {∫ [ N ] {b} dΩ}
A T
i
Ω
A T A
i
Ω
Τ
Inherent within this expression is the material constitutive law. This expression is computed for each and every
element in the element axes system. These contributions make up the nodal equilibrium equations at each and every
node. In order to assemble that, we again employ the principle of virtual displacements
∂W ∂{d}
T
= {f }
∂{U} ∂ U
{P} = [T]T {f }
or the principle of minimum potential energy
U= {d} [k ]{d}
1 T
(for linear elastic material)
2
δU = δ{d} [k ]{d} = δ{d} {f }
T T
δV = δ{d}T {f } − δW = {0}
∂V ∂{d}
T
∂W ∂{d}
T
= {f }
∂{U} ∂ U
{P} = [T ] T {f }
{P} = [TA]T{f}
where {P} is the nodal external force vector in global axes system, {f} is the elemental nodal force vector in the
element axes system and [T] is the transformation matrix. The vector {P} is thus the user-specified external loads
which are known. The transformation matrix [T] transforms the elemental contribution to the static equilibrium
equations such that compatibility is ensured. Also note that the displacement DOFs need also be transformed to
ensure compatibility
{d} = [TA]{U}
21
Structural Engineering Analysis and Design
Second Edition
We thus obtain the nodal static equilibrium equations in the global axes system
{P} = [T A ] T {f }
= [T A ] T {[∫ [B
Ω
]
{∫ [B ] {σ} dΩ}− {∫ [B ] [D ]{ε} dΩ}− {∫ [N] {b}dΩ}
] [D A ][B A ]dΩ [T A ]{U} +
A T
Ω
A T
i
Ω
A T A
i
Ω
Τ
[ ] [ ]
as K AE = [T A ] T k AE [T A ]
The first term on the RHS in the nodal static equilibrium equation expression is the global instantaneous stiffness
matrix. The second, third and fourth on the RHS are the fixed end forces in the global axes system. Note that the
only unknowns in this expression are the global nodal displacement DOFs {U}. Hence in a linear static solutions
scheme, this expression is solved for {U} after the essential boundary conditions are applied using a simultaneous
equation solving algorithm such as Gaussian Elimination.
The stiffness matrix prior to application of the essential boundary conditions is known as the initial stiffness matrix
whilst that after the application of the essential boundary conditions is known as the final stiffness matrix. The
essential boundary conditions may be zero (constraints) or non-zero (settlement or enforced displacement).
Applying zero essential boundary condition effectively refers to deleting both the row and column associated with
the constrained DOF. Applying a non-zero (settlement) essential boundary condition refers to deleting the row
but not the columns associated with the DOF with the settlement. The deletion of the row indicates that the
additional equation defining the DOF as an unknown is not utilized within the simultaneous solver. And by not
deleting the column, the remaining equations utilize the known displacement value, which becomes a part of the
fixed end forces when the term is brought over the loading side of the equations. Hence the specification of the
enforced displacement always requires the specification of a constraint to delete the row (SPC entry in
MSC.NASTRAN) and the specification of an enforced displacement to account for the known displacement value
for the column (SPCD entry in MSC.NASTRAN). For example if the following is the initial stiffness matrix where
f now denotes the fixed end forces
P1 f 1 K 11 K 12 K 13 K 14 U 1
P f K K 24 U 2
2 2 21 K 22 K 23
− =
P3 f 3 K 31 K 32 K 33 K 34 U 3
P4 f 4 K 41
K 42 K 43 K 44 U 4
and if the displacement U1 is a settlement or an enforced displacement, and U2 is a constraint, then the final
stiffness matrix
P1 f 1 K 11 K 12 K 13 K 14 U 1
P f K K 24 U 2
2 2 21 K 22 K 23
− =
P3 f 3 K 31 K 32 K 33 K 34 U 3
P4 f 4 K 41
K 42 K 43 K 44 U 4
giving
P3 f 3 K 31 U 1 K 33 K 34 U 3
− − =
P4 f 4 K 41 U 1 K 43 K 44 U 4
22
Structural Engineering Analysis and Design
Second Edition
Notice the settlements now contribute to the fixed end forces. This expression is solved simultaneously for the
unknown displacements U3 and U4. Solving these final stiffness static equilibrium equations simultaneously then,
effectively produces the deformed configuration of the structure.
The reactions (obtained for both zero and non-zero essential boundary conditions) are obtained from the deleted
rows of the initial stiffness matrix. For the above example, the reactions P1 (at the settlement) and P2 (at the zero
constraint) are
U1
P1 f 1 K 11 K 12 K 13 K 14 U 2
− =
P2 f 2 K 21 K 22 K 23 K 24 U 3
U 4
The reaction at the enforced displacement constrained should be checked after an analysis to ensure that the forces
required to effect this displacement are realistic.
Note that a static analysis can be solved with solely applied displacements, i.e. loads are not mandatory, as long as
the 6 rigid-body modes are constrained.
Again, it is important to grasp the concept that in geometrically linear (GL) global element stiffness formulations,
the transformation matrix [TA] is constant and thus independent of the global displacement vector {U}.
23
Structural Engineering Analysis and Design
Second Edition
The variation of displacement within a finite element can be presented as a function of the matrix of shape
functions [N] and the discrete nodal degree of freedom displacement vector {d} as
{y} = [N]{d}
Vector {y} describes the general displacement function (interpolation function) within the finite element, {d} is the
unknown nodal DOFs and [N] are their corresponding shape functions.
The general strain vector {ε} in terms of {d} can then be derived as the strain is some derivative function of the
displacement function and hence the nodal DOFs {d}
{ε} = [B]{d}
One of the differences with a nonlinear finite element stiffness formulation lies in the elemental strain expression
{ε} which is no longer linearly related to the nodal displacement DOFs {d}. Hence the strain matrix [B] is no
longer constant but instead a function of the nodal displacement vector {d}. This has further repercussions that will
become apparent.
The general stress vector {σ} can then be established in terms of the strains and hence be expressed in terms of the
DOFs {d} amongst other terms
The second term is due to the initial stresses (such as residual stresses within the element) and the third term is the
initial strains due to temperature shrinkage or lack of fit. For material nonlinearity, the constitutive matrix [D] is
also dependent upon the strains {ε} which in turn is dependent upon the nodal displacement vector {d}.
For a variational finite element, the equilibrium equation is thus derived by equating the external work to the
internal energy for a virtual displacement set or by minimizing the total potential energy (note that these theorems
are equally applicable for nonlinear systems as they are for linear systems)
V=U−W
δV = δU − δW = 0 for equilibrium
where {f} is the nodal force vector and {b} is the distributed body forces within the finite element. We have
however established that
{y} = [ N ]{d} hence {y}T = {d}T [ N ]T
{ε} = [B]{d} hence {ε}T = {d}T [B] T
{σ} = [D][B]{d} + {σ}i − [D]{ε}i
hence
The above expression is valid for any virtual displacement δ{d}. Choosing unity virtual displacements we arrive at
the elemental equilibrium equations
[
{f } = ∫Ω [B]T [D][B]dΩ {d} + ] {∫ [B] {σ} dΩ}− {∫ [B] [D]{ε} dΩ}− {∫ [N] {b}dΩ}
Ω
T
i
Ω
T
i
Ω
Τ
24
Structural Engineering Analysis and Design
Second Edition
This is the expression that is set up for each and every element. If the nodal force vector {f} is known, then the only
remaining unknown within these equations are the nodal displacement vector {d}. Note that we have defined [B]
the strain matrix, {σ}i = the initial stresses, [D] the material constitutive matrix, {ε}i = the initial strains, [N] =
shape functions matrix, {b} = the elemental (external) body loads and of course {d} = nodal element displacement
in element axes.
To summarize the terms in the above elemental static equilibrium equation expression
(i) the term on the LHS is the nodal force vector in element axis
(ii) the first term on the RHS is the instantaneous stiffness matrix
(iii) the second, third and fourth terms on the RHS are the so-called fixed end forces
We note that the choice of the shape functions [N] affects the accuracy of
This elemental static equilibrium expression seems exactly identical to that of the linear finite element. However,
there are inherent fundamental differences that must be realized, i.e.
(i) Firstly, the strain matrix [B] is no longer constant but instead dependent upon the nodal displacement
vector {d}, this being a characteristic of geometric nonlinearity.
(ii) Secondly, the constitutive matrix [D] is not constant but instead dependent upon the nodal
displacement vector {d}, this being a characteristic of material nonlinearity.
(iii) Thirdly, the body loading vector {b} is also dependent upon the nodal displacement vector {d}, this
also being a characteristic of geometric nonlinearity
An extremely fundamental concept to grasp is that as a result of the dependence of [B], [D] and {b} on {d}, we can
no longer solve for {d} using linear simultaneous equation solving algorithms. In the linear systems, we write the
expression
and provided we know the applied nodal forces {f}, which we do, we can solve for {d} as the fixed end forces are
independent of {d}. Now, in the nonlinear system, the equilibrium equations are nonlinear. We thus employ
another technique based on the tangent stiffness in order to solve the nonlinear equilibrium equations for {d}.
The stiffness matrix of the nonlinear system is no longer constant. By definition the stiffness of the system is
[k ] = ∂{f }
∂ d
For a linear finite element, since [B], [D] and {b} are independent of {d}, the second, third and fourth terms in the
elemental equilibrium equation does not feature in the stiffness expression which simply is
[k ] = [∫ [B] [D][B]dΩ]
Ω
T
On the other hand, for a nonlinear finite element, there are remaining terms when the second, third and fourth terms
of the elemental equilibrium equation are differentiated with respect to {d} to obtain the elemental tangent stiffness
expression as shown below.
25
Structural Engineering Analysis and Design
Second Edition
∂2V
[k T ] =
∂{d}∂ d
∂2U ∂2W
= −
∂{d}∂ d ∂{d}∂ d
∂{ε}
∂
= { }
∂2W
[B] T {σ} −
as [B] =
∂ d ∂{d}∂ d ∂ d
T ∂{σ} ∂[B]T 2
= [B] + {σ} − ∂ W using the derivative of product rule
∂ d ∂ d ∂{d}∂ d
T ∂{σ} ∂{ε} ∂[B]T 2
= [B] + {σ} − ∂ W using the chain rule
∂ ε ∂ d ∂ d ∂{d}∂ d
∂[B]T 2 ∂{σ} ∂{ε}
= [B] [D][B] +
T
{σ} − ∂ W as [D] = and [B] =
∂ d ∂{d}∂ d ∂ ε ∂ d
∂ ∂{ε} T ∂ 2 W ∂{ε}
= [B] [D][B] + {σ} − as [B] =
T
∂ d ∂ d ∂{d}∂ d ∂ d
∂ ε
2
2
= [B]T [D][B] + {σ} − ∂ W
∂{d}∂ d ∂{d}∂ d
[k T ] = [k E ] + [k G ]
[k E ] = ∫ [B]T [D][B]dΩ
Ω
∂2 ε ∂ 2 Wi ∂ 2 Wn
[k G ] = ∫ {{σ}n + {σ}i − [D]{ε}i }dΩ − − λ
Ω
∂{d}∂ d ∂{d}∂ d ∂{d}∂ d
{σ} n = [D][B]{d}
It is apparent now that the stiffness of the nonlinear finite element (and by stiffness we mean the variation of
resistance {f} with respect to the nodal displacement {d}) is not just dependent upon the instantaneous stiffness
[kE] but also includes second order internal strain energy and external work terms so-called the differential
stiffness terms due to the initial and nominal stresses within the finite element and the initial and nominal external
work done on the element.
To summarize, the primary differences with the nonlinear static stiffness formulation are
(i) the elemental strain {ε} have higher order (than linear) terms of the displacement DOFs {d} and
hence the strain matrix [B] is no longer constant but instead dependent upon the nodal displacement
vector {d}, this being a characteristic of geometric nonlinearity. Hence, firstly, the [B] term within
[kE] is dependent upon the state of deflections {d} instead of being constant with respect to {d}, and
secondly, the second derivative of the strain {ε} with respect to {d} is not zero thus causing the
addition of a geometric stiffness that is also a function of the prestress.
26
Structural Engineering Analysis and Design
Second Edition
(ii) the constitutive matrix [D] is not constant but instead dependent upon the nodal displacement vector
{d}, this being a characteristic of material nonlinearity.
(iii) the elemental loading vector {b} has higher order (than linear) terms of the displacement DOFs {d},
this also being a characteristic of geometric nonlinearity.
These differences produce stiffness terms that are not constant (or independent) with respect to {d} but instead
dependent upon {d}. The nonlinearity of the stiffness expression thus requires the evaluation of an appropriately
called tangent stiffness matrix.
Elements with GNL capabilities have thus the following additional capabilities: -
(i) The instantaneous stiffness matrix [kE] is dependent upon the deflected configuration. This is
because the strain-displacement relationship is nonlinear i.e. [B] would be a function of {d}.
Elements with the capability of nonlinear [B] matrices are termed large strain elements.
(ii) The geometric (or differential) stiffness matrix is dependent upon the state of stresses (or forces for
one-dimensional finite elements) as depicted by the {σn+σi+Dεi} term where {σi} is the initial
prestress and {εi} is the initial strain. Elements that are capable of modelling this effect are termed
large strain elements.
(iii) The geometric (or differential) stiffness matrix is dependent upon the second order variation of
initial and nominal work done with respect to the local element DOFs. There will be no
contribution from these terms if the applied external forces are work-conjugate with the element
DOFs. By work-conjugate, we mean that these second-order work terms will produce equivalent
nodal loading which are always in the direction of the local element nodal displacement DOFs {d}.
GNL elements capable of modelling the ‘follower force effect’ where the force direction is
dependent upon the deflected shape would make contributions to these terms as the equivalent
nodal force contributions are then not work-conjugate.
(i) The stress-strain relationships is nonlinear i.e. [D] would be a function of {ε} which in turn is a
function of the nodal displacement vector {d}
These GNL, MNL elements reduce to GL, ML elements if [B], [D] and {b} are constant in which case the stiffness
matrix will not be dependent upon the state of stress (or forces for one-dimensional finite elements) or the second
order work variation. The stiffness matrix then corresponds to that of the instantaneous stiffness matrix at the initial
undeflected configuration state, A.
∂2 ε ∂ 2 Wi ∂ 2 Wn
[k G ] = ∫ {{σ}n + {σ}i − [D]{ε}i }dΩ − − λ
Ω
∂{d}∂ d ∂{d}∂ d ∂{d}∂ d
{σ}n = [D][B]{d}
GL, ML elements
[ ]
[f ] = k AE {d}
[f ] = ∫ [B ] [D][B ]dΩ {d}
A T A
can be solved by linear simultaneous equation solving algorithms
Ω
27
Structural Engineering Analysis and Design
Second Edition
(i) Large strain effect – Non constant instantaneous stiffness due to non-constant [B] and the effect of
the geometric stiffness due to prestress and prestrain and nominal stress
(ii) The follower-force effect
(iii) Nonlinear stress-strain behavior
It is imperative to realize that the nonlinearity of [B] and the prominence of the element [kG] matrix both reduce
when the element length L (or other dimensions for two and three dimensional finite elements) reduces. This brings
us to a very important conclusion, i.e. that GL elements (small strain elements) can be used to approximate GNL
element (large strain element or hyperelastic elements) force-displacement response if a sufficient number of
elements are utilized to model a single structural member. This however does not mean that a GL analysis solution
scheme can be employed, instead a GNL solution technique must be used as the nonlinearities are accounted for in
the global behaviour and not the local element behaviour.
28
Structural Engineering Analysis and Design
Second Edition
The following describes the assembly of the element stiffness matrices towards the formulation of the global
system stiffness matrix by the use of the equilibrium, compatibility and constitutive laws. We have shown that the
elemental equilibrium equations are
[ ]
{f } = ∫Ω [B]T [D][B]dΩ {d} + {∫Ω [B]T {σ}i dΩ}− {∫Ω [B]T [D]{ε}i dΩ}− {∫Ω [ N]Τ {b} dΩ}
Inherent within this expression is the material constitutive law. This expression is computed for each and every
element in the element axes system. These contributions make up the nodal equilibrium equations at each and every
node. In order to assemble that, we again employ the principle of minimum potential energy
{P} = [T ]
T
{[∫ [B] [D][B]dΩ][T]{U} + {∫ [B] {σ} dΩ}− {∫ [B] [D]{ε} dΩ}− {∫ [N] {b}dΩ}}
Ω
T
Ω
T
i
Ω
T
i
Ω
Τ
Note that [T] is the transformation matrix between the element DOF system {d} to the global DOF system {U}.
This transformation matrix is dependent upon the global system parameters {U}. The local element axes system
changes with respect to the global element axes as it is dependent upon the position of the DOFs in the deflected
configuration, a so-called Eulerian system.
The equilibrium equations cannot be solved using a linear solution algorithm, as they are a set of nonlinear
equations, nonlinearly related to {U}, which are the only unknowns. Of course a nonlinear simultaneous algorithm
could in theory be applied, but this usually proves to be too computationally demanding, and inefficient. Hence,
solutions schemes almost always utilize the tangent stiffness matrix approach. The tangent stiffness is independent
of the unknowns {U}. Hence, for a small load or displacement step, the response of system can be approximated to
be linear with respect to {U}. In other words, for a small load or displacement step, the response of the system will
be linear and hence a linear simultaneous solution algorithm may be employed if the solution scheme is implicit.
Knowing the tangent stiffness matrix, we can write that for a small change in {U} (small displacement step) or for
a small change in {P} that
This is a linear relationship for the nonlinear problem. Hence, any solution scheme (static or dynamic) that
incorporates the variation of stiffness with deformation {U} i.e. nonlinear stiffnesses requires a stepping algorithm
in order to change the value of the tangent stiffness as the solution progresses. This is required as from the equation
29
Structural Engineering Analysis and Design
Second Edition
above, it is apparent that a particular linear equilibrium equation that represents the nonlinear problem is valid only
for a small load or displacement step. For completion, it is worth mentioning that when δ{P} is zero for a certain
small δ{U}, this means that [KT] is zero. Physically, this means that the system is at an unstable equilibrium state,
i.e. it is on the brink of buckling.
The expression for the tangent stiffness matrix in the global system will now be presented. We know from the
principle of minimum potential energy that a system is in equilibrium when the first order derivative of the total
potential energy with respect to {U} is zero. The tangent stiffness of the system is the second order derivative of
the total potential energy function with respect to {U}, i.e.
∂2V
[K T ] =
∂{U}∂ U
∂2U ∂2W
= −
∂{U}∂ U ∂{U}∂ U
∂2U ∂{P} ∂W
= − as {P} =
∂{U}∂ U ∂ U ∂{U}
But loads applied at nodes within a commercial finite element program are always in the direction of the global
system throughout the analysis. This means that the nodal loading is work-conjugate with the global displacement
vector {U} and thus there will be no second order variation in work. Thus the second term in the above expression
is zero. Note that the second order variation of work terms due to loading within the finite elements is taken into
account within the elemental formulations as described earlier. We thus arrive at
∂2U
[K T ] =
∂{U}∂ U
which states that the tangent stiffness matrix is equal to the second order variation of internal strain energy.
Performing some basic calculus
∂{d}
∂
[K T ] = {
[T]T {f } } as [T ] =
∂ U ∂ U
∂{f } ∂[T ]T
= [T ]T + {f } using the derivative of product rule
∂ U ∂ U
T ∂{f } ∂{d} ∂[T ]T
= [T ] + {f } using the chain rule
∂ d ∂ U ∂ U
∂[T ]T ∂{f } ∂{d}
= [T ] [k T ][T ] + {f } as [k T ] = and [T] =
T
∂ U ∂ d ∂ U
∂ ∂{d} T ∂{d}
= [T ] [k T ][T ] + {f } as [T ] =
T
∂ U ∂ U ∂ U
∂ d
2
= [T ] [k T ][T ] + {f }
T
∂{U}∂ U
30
Structural Engineering Analysis and Design
Second Edition
To summarize global tangent stiffness matrix replacing terms from the elemental tangent stiffness matrix
∂2 d
From above [K T ] = [T ]T [k T ][T] + {f }
∂{U}∂ U
but [ ] {
{f } = ∫Ω [B]T [D][B]dΩ {d} + ∫Ω [B]T {σ}i dΩ }− {∫ [B] [D]{ε} dΩ}− {∫ [N] {b}dΩ}
Ω
T
i
Ω
Τ
and [k T ] = [k E ] + [k G ]
where [k E ] = ∫ [B] [D][B]dΩ
T
∂2 ε ∂ 2 Wi ∂ 2 Wn
and [k G ] = ∫ {{σ}n
+ {σ}i − [D]{ε}i }dΩ − − λ
∂{d}∂ d
Ω ∂{d}∂ d ∂{d}∂ d
for which {σ} n = [D][B]{d}
Thus [K T ] = [K E ] + [K G ]
where [K E ] = [T ] [[k E ]][T ]
T
∂2 d
and [K G ] = [T]T [[k G ]][T] + {f }
∂{U}∂ U
∂2 ε ∂ 2 Wi ∂ 2 Wn
for which [k G ] = ∫ {{σ}n + {σ}i − [ D]{ε}i }dΩ − − λ
Ω
∂{d}∂ d ∂{d}∂ d ∂{d}∂ d
and {σ}n = [D][B]{d}
and [
{f } = ∫Ω [B]T [D][B]dΩ {d} + ] {∫ [B] {σ} dΩ}− {∫ [B] [D]{ε} dΩ}− {∫ [N] {b}dΩ}
Ω
T
i
Ω
T
i
Ω
Τ
Note that we have omitted the second order variation of work due to the external nodal loadings {P} from the
above tangent stiffness matrix expression, i.e.
∂2W ∂{P} ∂W
− = − as {P} =
∂{U}∂ U ∂ U ∂{U}
This is because, as mentioned, the externally applied loads at the nodes are in commercial codes always work-
conjugate with the nodal DOFs {U}, failing which the above term must be incorporated. The second order variation
of work due to loads applied on the finite elements is obviously still taken into account, namely in the terms
∂ 2 Wi ∂ 2 Wn
− − λ
∂{d}∂ d ∂{d}∂ d
Finally, the second-order work contribution of these elemental loadings to the internal energy of the global
formulation is
∂2 d
−
∂{U}∂ U
{∫ [N] {b}dΩ}
Ω
Τ
31
Structural Engineering Analysis and Design
Second Edition
Geometric nonlinearity can be broken down into just four component, firstly nonlinearity due to the
prominence of HIGHER ORDER ELEMENT STRAIN TERMS when local element deflections become
large, secondly nonlinearity due to HIGHER ORDER WORK TERMS (FOLLOWER FORCE EFFECT)
when global deflections become large, thirdly nonlinearity of the LOCAL TO GLOBAL
TRANSFORMATION MATRIX when global deflections become large and fourthly the nonlinearity in the
form of the global geometric stiffness matrix due to global PRESTRESS and large global displacements.
I. HIGHER ORDER ELEMENT STRAIN TERMS. Higher order element strain terms have two effects
upon the tangent stiffness matrix. Firstly, the local element instantaneous stiffness matrix [kE] becomes
dependent upon the local element deflected configuration and not the undeflected configuration.
[k E ] = ∫ [B]T [D ][B]dΩ
Ω
This is because the strain-displacement relationship is nonlinear i.e. [B] would be a function of {d} instead
of being constant and corresponding to the initial undeflected state [BA] as in a linear analysis. If {d} is
large, higher order terms in the strain-deflection {ε}=[B]{d} relationships become significant. Elemental
small displacement theory does not produce correct results. This occurs because the inherent instantaneous
stiffness of a structural element changes as its dimensions change as the analysis progresses. Elements with
the capability of nonlinear [B] matrices are termed large strain or hyperelastic elements. In large strains
elements, the strain matrix [B] varies with deflection {d}. Large strain effect occurs in metal forming,
rubber and elastomer applications. The material could be elastic with linear [D] but still it may
accommodate large strains within its elastic range, the higher order strains affecting its stiffness. Secondly,
higher order element strain terms cause nonlinearity of the local element tangent stiffness with respect to
local element deflections in the form of the element geometric stiffness due to prestress. In linear analysis,
elemental [kG] is not accounted for.
∂2 ε
[k G ] =
∫ {{σ}n + {σ}i − [D]{ε}i }dΩ
∂{d}∂ d
Ω
where {ε} = [B]{d}
{σ}n = [D][B]{d}
An important observation is that the nonlinearity of [B] and the prominence of the element [kG] matrix both
reduce when the element length L (or other dimensions for two and three dimensional finite elements)
reduces. This brings us to a very important conclusion, i.e. that GL elements (small strain elements) can be
used to approximate GNL element (large strain element or hyperelastic elements) force-displacement
response if a sufficient number of elements are utilized to model a single structural member. The smaller
dimensions of the small strain finite elements ensure that they are not distorted too much ensuring that the
internal strains are not too large so as to invalidate the linear [B] relationship. It is perfectly valid to allow
the small strain elements to undergo large total deformation as in rigid body motion, however its relative
deformation must be small for its linear [B] relationship to be valid. The nonlinearity of the system is then
accounted for in the global behaviour within the GNL solution technique and not the local element
behaviour.
II. HIGHER ORDER WORK TERMS (FOLLOWER FORCE EFFECT). The follower force effect is the
nonlinearity of the tangent stiffness with respect to deflections in the form of the geometric stiffness due to
external forces, also known as the follower force effect. The geometric (or differential) stiffness matrix is
dependent upon the second order variation of work done with respect to the local and global DOFs. The
stiffness of the structure is dependent upon the magnitude and direction of the external element loads,
32
Structural Engineering Analysis and Design
Second Edition
which again is dependent upon the current deflected configuration. This occurs when the applied force
depends on the deformation (so-called follower force) in the case of hydrostatic loads on submerged or
container structures, aerodynamic and hydrodynamic loads caused by the motion of aeriform and
hydroform fluids (wind loads, wave loads, drag forces). Element loads often tend to follow the normal to
the surface of the element. In linear analysis, these terms are not included.
∂ 2 Wi
−
∂{d}∂ d
∂ 2 Wn
− λ
∂{d}∂ d
∂2 d
= −
∂{U}∂ U
{∫ [N] {b}dΩ}
Ω
Τ
Nodal forces and accelerations are always in the direction of freedoms and hence are work conjugate and
hence linear analysis is sufficient to model these. Hence, there will be no contribution from these terms if
the applied external forces are work-conjugate with the element or global DOFs. GNL elements capable of
modelling the ‘follower force effect’ where the force direction is dependent upon the deflected shape would
make contributions to these terms as the force is then not work-conjugate.
∂2W ∂{P} ∂W
− = − as {P} =
∂{U}∂ U ∂ U ∂{U}
III. PRESTRESS. This is nonlinearity in the form of the global geometric stiffness matrix due to global
PRESTRESS and large global displacements.
∂2 d
{f }
∂{U}∂ U
where [ ]
{f } = ∫Ω [B]T [D][B]dΩ {d} + {∫Ω [B] T {σ}i dΩ}− {∫Ω [B]T [D]{ε}i dΩ}− {∫Ω [ N]Τ {b} dΩ}
Geometric stiffening or destiffening is due to element stress, prestress and prestrain. There is a contribution
to the geometric (or differential) stiffness matrix from the fact that an element would have internal stresses
and forces. The stiffness of the structure is dependent upon the current magnitude and nature of the internal
force actions within the structural element that is dependent upon the current deflected configuration. A
tensile load within a structural element increases the stiffness (increasing its natural frequency) as it tends
to return the element to its undeflected shape whilst a compressive load decreases the stiffness (decreasing
its natural frequency) as it tends to amplify the deflected shape. Hence the high bending stiffness in
prestressed cables where an unstressed cable will have negligible bending stiffness. Likewise is the
reduction in stiffness in a compressed column. Also, the gradual prestressing of a structural element such as
a thin plate that is restrained from longitudinal contraction and subject to bending causing the neutral
surface to stretch and generate significant tensile stresses, which serve to stiffen the plate in bending due to
the geometric stiffness. Of course, if the plate is not restrained, there will simply be longitudinal
contraction and no in plane stress will be generated, a linear analysis is then often sufficient. Clearly, this
longitudinal contraction when there are no restraints or the generation of in plane forces due to transverse
bending when there are restraints, is only observed in a nonlinear analysis, not a linear one. Hence, if there
is an element prestress {σi} or if the deflections {d} or {U} is sufficiently large to make the change in
element stress {σn} prominent in affecting the stiffness as the analysis progresses, the geometric (or
differential) stiffness matrix should be accounted for. In linear analysis, these terms are not included unless
special techniques such as P-∆ analyses are performed.
IV. LOCAL TO GLOBAL TRANSFORMATION MATRIX. This is the nonlinearity of the transformation
matrix [T] with respect to deflections {U}. If the deflections (U) are large, then it is appropriate to account
for the variation of the [T] matrix with {U}. Even if the elemental strains are small (which can be the case
if the finite element mesh is very fine), if the global deflections {U} are large, it is prudent to employ a
nonlinear solution scheme.
33
Structural Engineering Analysis and Design
Second Edition
[T] = ∂{d}
∂ U
In linear analysis [T] is constant and is based on the initial undeflected configuration, i.e. [TA]. In
geometrically nonlinear analysis, equilibrium and compatibility is satisfied in the deformed configuration
whilst in geometrically linear analysis, equilibrium and compatibility is satisfied in the undeformed
configuration. Further, the force transformation matrix is not the transpose of the displacement
transformation matrix in geometrically nonlinear analysis whilst it is in linear analysis. Thus, since
displacements are directly proportional to loads in linear analysis, results from different load cases can be
superimposed, whilst that cannot be done for nonlinear analysis. Accounting for the nonlinearity of [T] is
significant for large displacements and rotations in the case of cables, arches and thin plates. Consider a
thin plate subjected to transverse loads. Shell elements are based on classical bending theory, which
assumes that the transverse loads on a flat panel are resisted by bending alone. As a thin shell has little
bending resistance a linear analysis would then predict a very large displacement, much larger than the
limit of displacement/thickness ratios (often quoted from 0.3 to 1.0) for which the classical theory applies.
In reality, once some deformation has occurred a thin shell can resist loads by membrane tension, which is
very stiff compared with bending. Hence when we talk of small displacements for linear analysis to be
valid, it should be quoted in view of the thickness of the element, not the absolute displacement magnitude.
This is a geometric nonlinear effect where the load path and nature of deformation and stress field
literally changes with deflections, an effect that cannot be predicted by linear analysis. This does not occur
with curved panels such as cylinders under internal pressure, where the transverse load is carried by hoop
stresses, in which case a linear analysis may be sufficient.
In nonlinear material finite elements, the stress-strain relationships is nonlinear i.e. [D] would be a function of {ε}.
[B] [D][B]dΩ
∫
T
Ω
In linear analysis, the stress-strain relationship is linear.
Nonlinear analysis is also able to change the boundary conditions as the analysis progresses. Contact can be
simulated where two or more parts of the structure collide, slide and separate.
2.1.6 GL, ML and GNL, MNL Dynamic Finite Element Elemental and Global Formulation
Dynamic analysis formulations differ from static analysis by the fact that in general they require two additional
matrices in the mass [M] and the damping [C] matrices. Assumed element interpolation functions now define the
element mass, damping and stiffness matrices, i.e. employing the constitutive law. The dynamic global equilibrium
equations are assembled (equilibrium law) from element contributions transformed (compatibility law) as
appropriate. The general global dynamic equilibrium equations for nonlinear analyses are
34
Structural Engineering Analysis and Design
Second Edition
Methods of structural static analysis may or may not incorporate nonlinearity in the form of geometric nonlinearity
and/or material nonlinearity.
λ
ML, GL Static Analysis
MNL, GL Plastic
Collapse Analysis λP
35
Structural Engineering Analysis and Design
Second Edition
The modal properties of linear systems can be ascertained by either one of the following methods: -
(i) ML, GL real modal (eigenvalue) analysis SOL 103 to establish the modal frequencies, real modal
mass (and modal stiffness) and the real mode shapes
(ii) ML, GL complex modal (eigenvalue) analysis SOL 107 to establish the modal frequencies, modal
damping, complex modal mass (and complex modal stiffness) and the complex mode shapes
(iii) ML, GL time domain implicit (SOL 109, SOL 112 or SOL 129) or explicit (LS-DYNA) impulse
analysis to excite the modes of interest. The duration of the impulse must be sufficiently long (td/T1
> ~0.4) to excite the first fundamental mode, which is usually of concern. This would result in a
response that includes the first fundamental mode and most likely higher modes as well. The first
fundamental mode is readily ascertained from inspection of the response time history curve at any
node. Higher natural frequencies can also be ascertained by performing an FFT on the response
curve.
The methods of calculating forced dynamic response in MSC.NASTRAN are particularly of concern. Dynamic
forced response analyses can be performed in the frequency or the time domain.
It is important to understand the components of the result solved for by NASTRAN from the different forced
response solution schemes. In general, there are two components of response, i.e.
Methods of calculating the forced response generally depend on the nature of the excitation function. Dynamic
analysis can be performed in either the time domain or the frequency domain. Time domain dynamic analyses
include both transient and steady-state response, whilst frequency domain dynamic analysis only computes the
steady-state response. The types of excitations and the corresponding solution methods are: -
(i) ML, GL Deterministic periodic harmonic long duration excitations; The starting transient is
insignificant compared to the steady-state response. Forced response (steady-state) is performed in the
frequency domain using SOL 108 or SOL 111. Damping estimates ARE crucial.
36
Structural Engineering Analysis and Design
Second Edition
(ii) ML, GL Deterministic periodic non-harmonic long duration excitations; The starting transient is
insignificant compared to the steady-state response. A periodic (of period T) function that is not
necessarily harmonic can be expressed as a summation of an infinite number of sine and cosine terms,
i.e. a Fourier Series. Forced responses (steady-state) are performed in the frequency domain using SOL
108 or SOL 111 with these individual harmonics (with the correct amplitudes and phase differences) as
the excitations. The total steady-state response is thus the summation of the responses of the individual
harmonics. Damping estimates ARE crucial.
(iii) ML/MNL, GL/GNL Deterministic non-periodic short duration impulse (a.k.a. blast) excitations
with subsequent wave propagation; The starting transient is significant as is steady-state response.
Sources of excitation include the start up of rotating machines (not the continuous functioning) being
brought to speed passing through various frequencies, but not sustaining any as to cause steady-state
conditions. Other sources include sudden impulse (force applied is the rate of change of momentum of
relatively small mass compared to mass of structure) or blast excitations. Forced response (starting
transient and steady-state) is performed in the time domain using SOL 109, SOL 112 or SOL 129.
Material properties that are not only nonlinear with the level of load but also the rate of loading (strain-
rate) may have to be incorporated in these sudden impulse analyses. Damping estimates ARE NOT
crucial.
(iv) ML, GL Random stationary (and ergodic) long duration excitations; This is the random equivalent
of the deterministic periodic (harmonic or non-harmonic) long duration excitation. The starting
transient is insignificant compared to the steady-state response. Forced response (steady-state) is
performed in the frequency domain using random vibration analysis (with Fast Fourier Transforms
FFT) utilizing SOL 108 or SOL 111. Damping estimates ARE crucial. Forced response in the time
domain using deterministic methods may not be appropriate as they will be too computationally
expensive due to the long duration of the excitation and the maximum response will be governed by
steady-state conditions, not by the starting transient response.
(v) ML/MNL, GL/GNL Random non-stationary short duration impulse excitations; This is the
random equivalent of the deterministic non-periodic short duration impulse excitations. The starting
transient is significant as is the steady-state response. Earthquake signals can be said to fall in this
category. Forced responses (starting transient and steady-state) are performed in the time domain using
SOL 109, SOL 112 or SOL 129 with a set of deterministic transient excitations generated from the
non-stationary random excitations and the results enveloped. Alternatively, forced response (starting
transient and steady-state) is performed crudely in the time domain using response (shock) spectrum
analysis. Damping estimates ARE still quite crucial. Steady-state forced response in the frequency
domain may not be appropriate since the duration of excitation may not be long enough for the
response of the structure to reach steady-state conditions.
(vi) MNL, GNL Projectile (a.k.a. impact) excitations with subsequent wave propagation; Forced
response (starting transient and steady-state) can only be performed in the time domain using an
explicit (implicit would be too expensive because of the high nonlinearity) nonlinear time integration
scheme. Sudden impulse analyses involve the specification of an initial velocity to the impacting
particle placed very close to the impacted structure. The particle will travel at this constant initial
velocity (since no other forces, retarding or accelerating) until impacting the structure whereby a
certain percentage of the mass of the structure and the mass of the particle will interact to cause a
change of momentum, i.e. an impulse. Material properties that are not only nonlinear with the level of
load but also the rate of loading (strain-rate) will have to be incorporated in these sudden impact
analyses. Damping estimates ARE NOT crucial.
(vii) MNL, GNL Brittle snap or redundancy check; These include the analysis of sudden fracture of
structural members from the static equilibrium configuration such as the sudden snapping of a
prestressed structural cable, and evaluates the redundancy available within the structure. Forced
37
Structural Engineering Analysis and Design
Second Edition
response (starting transient and steady-state) can only be performed in the time domain using an
explicit (implicit would be too expensive because of the high nonlinearity) nonlinear time integration
scheme. This analysis always involves a two-stage procedure. The first stage involves a static solution
by a linear static method or a nonlinear static method (either Newton’s tracing the equilibrium path
SOL 106, an implicit dynamic relaxation by SOL 129 or an explicit dynamic relaxation method LS-
DYNA) in order to obtain the deflected configuration and the stiffness of the structure in the deflected
configuration KT by the prestress (prestressing of structural members such as cables contribute greatly
to a large change in stiffness within the static solution and hence will require a nonlinear static method)
and gravity. The second stage involves a restart into a nonlinear transient dynamic solution scheme
with no additional dynamic excitation but with a change in the structure (in the deletion of a member
that fails in brittle fashion simulating a redundancy check) or a change in boundary condition (to
simulate the effect of the loss of a support or anchor attached to a prestressed cable etc.). An example
would be the gradual prestressing of structural cable elements within a suspension bridge or a cable
prestressed tower until a static solution (by SOL 106, implicit SOL 129 or explicit LS-DYNA) is
achieved. The cables can be prestressed using a gradual temperature load case (or a gradual enforced
displacement on the cable anchorage points) plus gravitational loads until the correct level of prestress
is achieved (by SOL 106, implicit SOL 129 or explicit LS-DYNA). Then with a restart into an explicit
transient dynamic scheme of LS-DYNA, the boundary condition supporting the cable is released or a
cable element is deleted, causing the structure to experience an out-of-balance of forces and hence
vibrate to a new static equilibrium configuration. Damping estimates ARE NOT crucial.
38
Structural Engineering Analysis and Design
Second Edition
A dynamic analysis is always performed about the static equilibrium position. That is to say, the structure should be
in static equilibrium before dynamic loads are applied. To obtain the total response, a few approaches may be used.
In the first approach, valid for both LINEAR AND NONLINEAR TIME DOMAIN solutions, the static
response must be added to the dynamic response if the dynamic analysis is performed about the initial undeflected
(by the static loads) state with only the dynamic loads applied, hence causing the dynamic response to be measured
relative to the static equilibrium position. This is because if the response is taken to be measured from this static
equilibrium position, it can be shown that the static loads do not feature in the dynamic equation of motion. This
enables the exclusion of static loads in the dynamic analysis. However, since the dynamic response is measured
from the static equilibrium position, the total response = the dynamic response + the static response to static
loads.
In the second approach, valid for both LINEAR AND NONLINEAR TIME DOMAIN solutions, if the
dynamic analysis is performed with the deflected static shape as initial input and the static loads maintained
throughout the dynamic excitations, the total or absolute response (static and dynamic) is obtained straight away
from the dynamic analysis. Hence total response = dynamic response (which already includes the static
response to static loads).
In the third approach, valid for LINEAR FREQUENCY DOMAIN solutions, not only that the static response
has to be added separately, but also the mean of the dynamic excitation has also got to be added separately as a
static response. This is because for both deterministic periodic excitations and random PSD excitations the mean of
the dynamically applied force is not included in the dynamic excitations. Hence the total response = static
response to mean of dynamic excitation + dynamic response + static response to static loads.
Analyses Sequences
Static Analysis Dynamic Analysis Remark
Modal analysis about static equilibrium position.
GL modal analysis
None Performed if both static and dynamic displacements
with [KEA]
are small.
GL Modal Analysis
39
Structural Engineering Analysis and Design
Second Edition
DYNA) analysis to response analysis are large and dynamic displacements are small. The
establish [KTA] at position with [KTA] (no dynamic response is measured from the static
of static equilibrium A. static loads, only equilibrium position. Hence, to ascertain the total
Approximate [KTA] can transient loads) response, the static response to static loads must be
be obtained by repetitive superposed.
P-∆ static analysis.
GNL (SOL 106, implicit
GL transient
relaxation SOL 129 or
response analysis
explicit relaxation LS-
with [KTA] and
DYNA) analysis to
initial dynamic Forced transient response analysis about initial
establish [KTA] at position
displacements as undeflected by static loads position i.e. the absolute
of static equilibrium A
the converged static response. Performed if static displacements are large
and the converged static
displacements and dynamic displacements are small. The response is
displacements.
(with constant with the absolute response, i.e. includes both the static and
Approximate [KTA] and
time equivalent dynamic response.
the converged static
dynamic static
displacements can also be
loads, and the
obtained by repetitive P-
transient loads)
∆ static analysis.
40
Structural Engineering Analysis and Design
Second Edition
Note that when restarting a transient analysis from a nonlinear static analysis (SOL 106, implicit relaxation SOL
129 or explicit relaxation LS-DYNA), the static loads which have been applied must be maintained throughout the
transient phase in order to preserve the correct stiffness and the correct static equilibrium at the deflected static
equilibrium position. Hence, when a transient phase dynamic analysis follows a dynamic relaxation analysis, it is
prudent to define two curves for the static loads or prescribed motions, one ramped from zero to the static value in
the dynamic relaxation phase and the other maintained at the same static throughout the transient dynamic analysis
phase. Transient analysis is usually performed after a static solution has been achieved either by SOL 106, implicit
relaxation SOL 129 or explicit relaxation LS-DYNA. The only benefit of performing dynamic relaxation prior to a
transient analysis is that the structure would reach its static equilibrium position quicker before the transient phase
takes place. In dynamic relaxation, the damping applied very large, in the order of about 5% of critical, whereas in
the transient phase, the structure is very much under-damped. But even if the transient analysis were to be
performed without dynamic relaxation, the structure would eventually vibrate about its static equilibrium position,
albeit in a while as long as the static gravitational loads are applied. Hence of course, if dynamic relaxation is
employed, after the dynamic relaxation phase, the static loads should be maintained throughout the transient
dynamic analysis phase. It is however, theoretically incorrect to vibrate the structure from an unstressed position as
in reality, the structure would be in its static equilibrium position before being subjected to transient dynamic
vibration. In practice however, it can be argued that a dynamic analysis can be done without a static solution if
negligible change in stiffness occurs from the static deflections (valid if static deflections are small), of course so
long as it is clear that the total response must then be the addition of the static response and the dynamic response,
both performed separately from the initial undeflected configuration.
41
Structural Engineering Analysis and Design
Second Edition
We have shown that the nodal static equilibrium equations can be written as a set of linear simultaneous equations
[ ] {∫ [B
{P} = K AE {U} + [T A ]T A T
}
] {σ}i dΩ − [T A ] T {∫ [B ] [D ]{ε} dΩ}− [T ] {∫ [N] {b}dΩ}
A T A
i
A T Τ
] [∫ [B ] [ D ][B ]dΩ][T ]
Ω Ω Ω
where [K ] = [T ] [k ][T
A
E
A T A
E
A
] = [T A T A T A A A
Ω
Geometrically linear stiffness response represents only a tangential approximation to the fundamental path at the
initial unloaded state. Hence, the results are valid in a range of small displacements and loading. Tangential
accuracy of linear analysis depends upon the magnitude of deflection, the component forces and the applied
loading. The geometrically linear equilibrium equations are solved for the unknown displacements {U}.
[ ]
Linear static K AE {U} = {P} + {Fixed End Forces}
[ ]
Linear static P - Delta K TA {U} = {P} + {Fixed End Forces}
The linear static analysis is based on only the instantaneous stiffness matrix [KE] at the initial undeflected stage
[KEA]. The linear static P-∆ analysis is based on the instantaneous and geometric (or differential) stiffness at the
initial undeflected stage i.e. taking in to account the initial prestress and initial external load on the stiffness of the
structure. The linear equations are solved simultaneous by decomposing [K] to triangular matrices, and then
performing forward-backward substitution to obtain {U}. This is called the LU-decomposition method of solving
linear simultaneous equations.
42
Structural Engineering Analysis and Design
Second Edition
The geometric stiffness or P-∆ concept will be illustrated for a simple system. A parameter that describes the
prominence of the geometric stiffness for a particular system and level of loading will then be presented.
Let us assume that deflections are small (hence geometrically linear analysis) hence the rotation of the column
segments can be described and any change in height can be neglected. Lateral equilibrium at the central pin shows
that
Note that the spring force depends only on the deflection x, while the lateral force generated by the component of N
depends on the deflection plus the initial imperfection, x + x0. Rearranging so that we have the form Stiffness x
Displacement = Applied Force, we get
i.e.
We can see that the effect of the force N acting on this system can be seen as being like an additional stiffness term
2N/L, i.e. the geometric (aka differential) stiffness. The effect of the axial force is equivalent to an additional
stiffness of value 2N / L acting laterally at the pin location. If the force is tensile (i.e. positive) then the geometric
stiffness will act to increase the overall stiffness of the system. The effects on the system tend to be beneficial, the
lateral deflection and hence the restraining force in the spring are reduced. Some practical examples where this type
of stiffness is important are in the strings of musical instruments where the tension is adjusted to give the required
natural frequency, in the blades of jet engines where the vibration characteristics are improved and the root bending
1
DALLARD, PAT. Buckling – An Approach Based on Geometric Stiffness and Eigen Analysis. Ove Arup & Partners
International Ltd., London, June 1998.
43
Structural Engineering Analysis and Design
Second Edition
moments reduced by the tension stiffening arising from rotation and in the antisymmetric response of unbraced
cable trusses and cable nets where the ‘real’ (i.e. instantaneous, not geometric) stiffness may be negligible. On the
other hand, if it is compressive then the term will be negative and the system stiffness will decrease. The effects on
the system tend to be detrimental, the lateral deflection and hence the restraining force in the spring are increased.
To reiterate, if N is compressive then the geometric stiffness is negative and x (the deflection) will be greater than it
would be otherwise. If x is greater then the force in the spring will be greater. If we were designing that spring then
we would need a stronger one than we first suspected. Therefore the effect of negative geometric stiffness (also
known as the P-∆ effect) can be significant in design in some circumstances.
The question is, when will this negative geometric stiffness effect be significant and when can we ignore it? One
good answer to this question is to examine the dimensionless parameter λcr (also known as the critical linear elastic
buckling load factor of a system).
λcr indicates the level of significance of the negative geometric stiffness (P-∆ effects) in a structure. Finding the
value of λcr is effectively a linearized buckling analysis (SOL 105). Note that λcr refers to the critical linear elastic
buckling load factor and not to the slenderness parameter associated with member elements such as columns and
beams, which is also normally referred to by λ. In this simple pin jointed system example however, the buckling
factor can easily be derived by considering the condition at which the axial compression is so great that the lateral
stiffness of the system is zero.
i.e.
The λcr expression is independent of the lateral load and the initial lateral imperfections. Hence, in the
absence of geometric stiffness considerations, there would be an abrupt bifurcation in the behavior of the system
at the critical load. It is the deflection expression, presented hereafter, that the effects of the geometric stiffness
considerations are seen as causing progressive amplification to lateral loads and initial lateral imperfections as the
compressive load is increased until the critical load. Expressing the deflection expression in terms of Ncr
Expressing the deflection equation again, but now in terms of λcr and the deflection without geometric stiffness
considerations (i.e. N = 0)
we have thus the deflection expression a function of λcr, the initial imperfection x0 and xN=0
Deflection Expression
This expression consists of two terms, the first being the amplified lateral response to lateral load when there is
the geometric stiffness consideration and the second being the amplified (addition to the) lateral response to
an initial imperfection when there is the geometric stiffness consideration. We say ‘addition to the’ lateral
response for the second term because the deflection x is measured relative to the initial imperfection and not
from the initial undeflected configuration. If there were no geometric stiffness consideration, there would be an
abrupt bifurcation in the behavior of the system at the critical load Ncr. The effect of the geometric stiffness
consideration are seen as causing progressive amplification to the lateral loads and initial lateral imperfections as
the compressive load is increased until the critical load. It has the effect of amplifying the lateral deflection and
44
Structural Engineering Analysis and Design
Second Edition
hence the spring force, or more generally, the forces in the elements resisting the lateral load. This is the P-∆
effect. The effects are apparent in practice. With the geometric stiffness consideration (and the existence of
lateral loads and/or initial lateral imperfections), when a slender element is subject to axial compression, it tends
to gradually bend before reaching the point of elastic instability. With the geometric stiffness consideration (and
the existence of lateral loads and/or initial lateral imperfections), if a beam is subject to major axis bending,
there is a tendency for the beam to twist and bend about its minor axis before lateral torsional buckling eventually
occurs.
We shall investigate the amplified lateral displacement to lateral loads when there is the geometric stiffness
consideration, which was shown to be
The definition of λcr is handy because it means that if λcr is a positive number greater than 1, then you have a
structure that is stable, but that the structural response will be amplified by negative geometric stiffness effects. If 0
< λcr < 1, then the system is unstable. And if λcr is negative then the structural response will be attenuated by
positive geometric effects. Here are some values of λcr and their corresponding amplification values.
We can see that when λcr > 10 the effects of negative geometric stiffness (P-∆ effects) are so small as to be
insignificant to an engineering level of accuracy (accurate within 10%), whereas as they approach 1 they are very
significant indeed. This then is the test that we were looking for. We can use the value of λcr to decide whether we
need to take further account of negative geometric stiffness in our design and analysis. Generally,
If λcr > 10 then P-∆ effects are insignificant and can be neglected – Perform linear analysis
If 4 < λcr < 10, P-∆ effects (i.e. the geometric stiffness) should be incorporated – Perform P-∆ analysis
If λcr < 4, a second order nonlinear analysis should be undertaken – Perform GNL SOL 106 analysis
If λcr is 1 or less (but still positive) then amplification is infinite and some significant changes must be made to the
design! It would be a mistake to be too scared of low λcr values. A very ordinary column in a building can have a
λcr of < 1.6, and quite ordinary bracing systems for office buildings can have a λcr of < 4. But conversely it is not
something that we can ignore and if response calculations are intended, a GNL SOL 106 must be performed.
The amplified (addition to the) lateral displacement to an initial imperfection when there is the geometric
stiffness consideration was shown to be
45
Structural Engineering Analysis and Design
Second Edition
When λcr is large the response will be small, when λcr is small, but >1 the response will be large, and when λcr = 1
the response becomes infinite, in a similar way to the amplified lateral displacement to lateral loads due to the
geometric stiffness consideration.
The geometric stiffness equation for a 2D bar element is presented. Assume the bar has an axial force N (tension +)
and length L, and use the local directions shown in the following diagram.
The active terms relate the geometric stiffening forces to the products of member forces and deflections. The terms
in the instantaneous stiffness matrix KE depend on the Young’s modulus and Poisson’s ratio of the material, and the
geometry of the element (the geometry of a beam element, for example, would include both the cross sectional
properties and the length of the element). In contrast, the terms in the geometric stiffness matrix KG depend
(linearly) on the forces in the element and the element geometry. The variation of the KG matrix with element
forces means that any analysis using a KG matrix needs to be carried out in two stages. In the first stage the problem
is solved ignoring the effects of geometric stiffening, establishing a set of element forces. These element forces are
used in the second stage to construct the KG matrix. An important consequence of the variation of the KG matrix
with element forces is that the KG matrix for one load case will differ from that for another. Hence while we can
always refer to the instantaneous stiffness matrix as KE, we need to identify the geometric stiffness matrix by
reference to the load case m for which it was derived, say KGm. The KG matrix is symmetric with connectivity
similar to the KE matrix.
In conclusion, the P-∆ effect is the reduction of lateral stiffness when the element is subjected to axial
compression such that there is an amplified lateral response to lateral loads and an amplified (addition to
the) lateral response to initial lateral imperfections (and subsequently the response of the elements providing
lateral restraint) according to
1
1 λ cr λ cr 1
x= x + x i.e. x= x N =0 + x0
1 N =0 1 0 λ cr − 1 λ cr − 1
1− 1−
λ cr λ cr
We have said that the deflection x is measured relative to the initial imperfection x0. If instead we present the
deflection relative to the initial undeflected configuration, then we need to add x0. Hence
λ cr 1 λ cr 1 λ −1 λ cr λ cr
x + x0 = x N =0 + x0 + x0 = x N =0 + x 0 + cr x0 = x N =0 + x0
λ cr − 1 λ cr − 1 λ cr − 1 λ cr − 1 λ cr − 1 λ cr − 1 λ cr − 1
λ cr
x + x0 = (x N = 0 + x 0 )
λ cr − 1
46
Structural Engineering Analysis and Design
Second Edition
Thus the deflection (measured from the initial undeflected configuration) to lateral loads and initial lateral
imperfections is amplified by λcr / (λcr – 1). This is a very important formula and can be applied
approximately to all other general cases.
To reiterate this rather important point, the above amplification λcr / (λcr – 1) was derived for a pin jointed system.
It can be shown that the same amplification results for the deflection and stress effects on other general systems.
For example, a strut with an initial imperfection x0 would have a transverse deflection distribution according to
x0sin(πx/L). With an axial load N, the transverse deflection would be x0sin(πx/L) [λcr / (λcr – 1)] where the buckling
load is the Euler load, i.e. π2EI/L2. It follows that the bending moment distribution would be Nx0sin(πx/L) [λcr / (λcr
– 1)]. For another example, a beam with a UDL would have a central deflection of 5ωL4/384EI and bending
moment of ωL2/8. A good approximation to the central deflection of the beam-column with an axial load of N is
(5ωL4/384EI) [λcr / (λcr – 1)] where the buckling load is the Euler load, i.e. π2EI/L2. It follows that the central
bending moment is approximately (ωL2/8) [λcr / (λcr – 1)].
The above proceedings described P-∆ analysis as a procedure by which the lateral stiffness is altered when a
member is subject to an axial force, hence causing more lateral deflection and thus a greater bending moment in the
member. An analogous way of looking at this is to consider an axial strut also subject to a lateral force. The lateral
force causes a lateral deflection, ∆. On application of the axial force P, a bending moment equal to P.∆ will be
induced in the member.
47
Structural Engineering Analysis and Design
Second Edition
3.1.3 Performing P-∆ (KGA From KEA) Linear Static Analysis – Direct Approach
This equation is directly analogous to the equilibrium equation derived for the simple pin jointed system above.
[K A
E ] [ ]
+ K GA {U} = {P} + {Fixed End Forces} − K GA {U 0 }
Note that {U0} is a vector of initial imperfections. Because {U0} is difficult and unintuitive to define, the modal
approach presented in Section 3.1.4 should be employed to perform the P-∆ analysis.
To perform a P-∆ analysis, first the value of λcr is established to decide whether and how account is made of the
negative geometric stiffness.
If λcr > 10 then P-∆ effects are unlikely to be significant and can be neglected
If 4 < λcr < 10, P-∆ effects (i.e. the geometric stiffness) should be incorporated
If λcr < 4, a second order nonlinear analysis (GNL analysis SOL 106) should be undertaken
Hence, only if λcr > 4 can we employ the P-∆ method for response computations. Any less, and a SOL 106 must
be undertaken. To perform the P-∆ analysis, two linear static analyses are required. First, a linear static analysis is
performed with the (ultimate limit state) loads m upon the instantaneous stiffness KEA to assess the distribution of
axial forces in the structure. These axial forces are used to generate the geometric stiffness matrix for that load case
m, KGmA which is combined with KEA to form the tangent stiffness matrix KTmA. A second linear analysis is then
performed with the (ultimate limit state) loads upon the tangent stiffness KTmA to generate the P-∆ response.
{U m } = [K AE + K Gm
A
] {{P
−1
m } + {Fixed End Forces}m [
− K Gm
A
]
{U 0 } }
Note that when we are doing an ultimate limit state design, as we generally are, the axial forces that cause the P-∆
amplification need to be ultimate axial forces, otherwise we will underestimate the amplification. For this reason
the loading on the structure needs to be the ultimate loading. And the results will be ultimate forces. You can’t
apply characteristic loads, run the P-∆ analysis, and then factor the loads and responses up, because the
amplification effect is nonlinear, as observed on the P-∆ curve below, i.e. if the initial compressive loads m
were higher, then the geometric stiffness will be larger and the response will be greater. Of course, the initial
loads m should be too great as to cause λcr to be less than 4. The down side of this is that each P-∆ case has to be
run as an individual analysis and you can’t combine P-∆ results in the conventional way, though you can of
course envelope them.
λ ML, GL Static Analysis
48
Structural Engineering Analysis and Design
Second Edition
The P-∆ method can be used to evaluate the fundamental buckling load factor. If the loads m were made too
high, the geometric stiffness may become greater than the instantaneous stiffness, and the equations cannot be
solved. Linear elastic buckling would have occurred. This means that we could in theory gradually increase the
load m, run the P-∆ analysis involving the two passes each time, until the equations cannot be solved. The load at
this stage is the lowest elastic buckling load. Performing this traces the above P-∆ equilibrium path. Note that the
line is curved, but not flat at the onset of buckling as it is not a nonlinear method. Response calculations near the
buckling load may not be accurate as λcr would be less than 4. It is much faster (and far more efficient in design) to
evaluate the buckling load factor by performing an eigenvalue analysis (SOL 105). Moreover, the buckling mode
shape will be clearly visible.
The P-∆ method may NOT be accurate in evaluating the deflection response just before the onset of the
fundamental buckling mode.
[K A
E ] [ ]
+ K GA {U} = {Pcr } + {Fixed End Forces} − K GA {U 0 }
P-∆ analysis is only accurate when λ is large, i.e. λ > 4. Hence, if the loads p0 were scaled to just below λp0, and a
P-∆ analysis undertaken, the displacement response may be inaccurate because now the load factor on λp0 is very
close to unity. A nonlinear static and buckling analysis SOL 106 is recommended.
49
Structural Engineering Analysis and Design
Second Edition
3.1.4 Performing P-∆ (KGA From KEA) Linear Static Analysis – Modal Approach; And Hence P-∆ Based
Buckling 2
3.1.4.1 Objectives
Finding the P-∆ response by solving the P-∆ equation directly is presented in Section 3.1.3. While this method
certainly works, a more intuitive modal approach may be desirable. The advantages of the modal approach are that
(i) it helps to determine the most critical load combination to maximize the P-∆ effects
(ii) it also allows for the imperfections (modeled to be proportional to the mode shape) to be
incorporated to maximize the response and hence determine the elastic buckling capacity based on
a P-∆ approach as is consistent with the BS 5950 code based Perry-Robertson method.
The basis of the modal approach is the ability of the buckling mode shapes to orthogonalize the P-∆ static
equilibrium equation into a set of uncoupled modal equilibrium equations, which can be solved independently for
the modal responses. These modal responses can then be transformed into the physical coordinates in order to
obtain the physical response. In doing so, of course there will be a certain degree of modal truncation. To obtain the
same answer as that of the direct approach (without imperfections), ALL the modes (equal to the number of DOFs)
need to be employed. Since it is difficult (and unnecessary) to determine which modes really contribute to the static
response, we shall employ the direct approach anyway. But with the addition here, that we include the response
due to the imperfections as well. The imperfections are assumed to be proportional to the mode shape. In fact, the
imperfections that really need to be incorporated are that of only the lower modes (λcr < 10). Thus, the modal
approach here is not intended as an alternative to the direct approach, but simply as an addition to incorporate the
(modal) imperfections for a P-∆ based buckling capacity prediction in line with the code based Perry-
Robertson method and also to aid the selection of the critical load case.
A linear static analysis is first performed with load case m upon the instantaneous stiffness KEA in order to generate
the geometric stiffness matrix for the particular load case KGm. The eigenvalue problem is
Note that the eigenvalue problem has to be associated with a particular load case m as the geometric stiffness
matrix is dependent upon the particular load case. The eigenvalue problem is solved for the critical load factors and
associated mode shapes.
Solving the eigenvalue problem does not however produce a physical response. For this we consider the static
equilibrium equation.
2
DALLARD, PAT. Buckling – An Approach Based on Geometric Stiffness and Eigen Analysis. Ove Arup & Partners
International Ltd., London, June 1998.
50
Structural Engineering Analysis and Design
Second Edition
Note that the initial imperfections is really just an additional load and is probably more illustrated if the associated
terms were brought onto the RHS of the equation as –KGmx0. This equation can be solved directly as performed in
Section 3.1.3. But this provides little insight into the buckled mode and hence the critical loading and pattern of
imperfection that should be considered in design. As mentioned, the following eigenvalue approach is much more
insightful in determining the response (deflections and forces). Representing the deflections as the summation of
the responses in the individual modes
and the initial imperfections as the summation of imperfections in the individual modes (hence assuming that the
shape of the imperfections is the same as the shape of the buckled mode shapes)
On substitution,
Premultiplying by the transpose of the mode shape matrix, rightly treating the imperfections as additional loads
Both KE and KGm will be diagonalized to give the modal stiffness and modal geometric stiffness
This is now a set of independent equations. Consider the ith equation as typical
The relationship between the modal stiffness, the modal geometric stiffness and the critical load factor λcri is
Also, from the modal response expression, the modal response ignoring the geometric stiffening effects is seen to
be the modal load {φi}T{f} divided by the modal stiffness
51
Structural Engineering Analysis and Design
Second Edition
This modal deflection expression is analogous to the deflection expression derived in Section 3.1.2 for the simple
pin-jointed column. Again, when there is geometric stiffness considerations, there is an amplified modal
response to modal loads and an amplified (addition to the) modal response to initial modal imperfections
(and subsequently the response of the elements providing restraint to the modal response).
A consequence of the modal nature of equation of the modal deflection expression is that the overall response
amplifier is difficult to predict. The amplifier 1/(1-1/λcr1) corresponding to the lowest buckling mode only applies
to that part of the loading which contributes to the modal force {φ1}T{f}, different amplifiers apply to the other
components of load. Hence it is unlikely that the total response will be amplified by 1/(1-1/λcr1). To illustrate this
consider a simple pin ended column subject to axial load and distributed lateral load. The 1/(1-1/λcr1) amplifier
would apply overall if the lateral load distribution followed the sinusoidal shape of the first buckling mode. This
pattern of loading would result in zero modal forces in all modes except the first. If the lateral load distribution was
uniform, say, it would generate modal forces in all odd numbered modes, each of which would have a different
modal amplifier, and would result in a different overall amplifier.
And by employing ALL the modes, the exact same response as that from the direct approach (without
imperfections) will be obtained. But that is not the purpose of our exercise here. We are going to employ the direct
approach anyway, this exercise is simply to aid the selection of the critical load case and for the modelling of the
(modal) imperfections.
The loading used to design a structure prone to buckling must be selected with care. This is done in order to obtain
the lowest possible load factor. Greater understanding of what loading may be critical can be gained from the
modal deflection equation. It is useful to consider the loading on a structure as consisting of two parts applied
simultaneously
(a) loading which determines the value of the amplifier (via λcri) but does not contribute to the linear
modal response, qi’
(b) loading which determines the value of the linear modal response qi’ (via {φi}T{f}) but does not affect
the amplifier (via λcri)
In a column, the axial load determines the value of the amplifier but does not contribute to the linear bending, while
the lateral load determines the linear bending but does not affect the amplifier. Consider an arch prone to sway
buckling and subject to asymmetric loading
52
Structural Engineering Analysis and Design
Second Edition
Divide the loading into symmetric funicular and antisymmetric non-funicular components.
The funicular loading determines the axial force in the arch and hence the amplifier, but does not contribute to the
linear bending. The antisymmetric loading determines the linear bending but does not affect the amplifier.
In the design of buckling sensitive structures, consideration needs to be given to how these two types of loading
vary against each other. The greatest overall response is likely to occur when the product of the linear
response and the amplification factor is maximized.
Imperfections corresponding to a particular mode shape cause response in that mode shape and not in any other.
The modal response (in modal coordinates) due to imperfections is 1/(λcri–1) times the magnitude of the modal
imperfection, q0i. Thus in physical coordinates, the modal response due to imperfections will be (1/(λcri–1))q0i{φ}i.
Hence, the additional element forces due to imperfections can be calculated in a post-processing operation, using
the additional modal responses due to imperfections simply to scale the modal element forces. It is usually
sufficient to consider imperfections mode by mode, rather than combined. The value of q0i should be selected to
cover the effects of all imperfections, including residual stresses and tolerance.
Now, from the theoretical derivation of the Perry-Robertson buckling capacity formula, the theoretical Perry
imperfection factor η is obtained
Ax 0 x y x y x L y
η= = 02 = 0 = 0
Z r r r L r r
where x0 is the initial imperfection. The Perry factor η is the imperfection factor which is dependent upon the
initial eccentricity, x0 and also the cross sectional properties. But BS 5950-Part 1:2000 goes a step further and
requires an increase in the Perry factor due to residual stresses from rolling and welding. It thus states that the Perry
factor should be
a (L e − L o )
η= a = Robertson constant
1000 ry
The Robertson constant, a divided by 1000, i.e. a/1000 is equivalent to (x0/L)(y/r) and includes the initial
imperfection, the section shape and also residual stresses. On equivalence of the two expressions for the Perry
factor, an equivalent initial imperfection x0 (or rather q0i here) can be obtained as
x0 ry a
q 0i = where x 0 = Ld
xg y 1000
The value of the initial imperfection x0 is purposely derived to be perfectly consistent with the BS 5950-Part 1:2000
Perry-Robertson formula. We can get values for x0 that depend on the strut curve (Table 23 in BS 5950-Part
1:2000), the length of the buckled waveform Ld, and a single geometric property (ry/y) of the section that is
buckling. The table below shows some appropriate values of x0 to use in common cases.
53
Structural Engineering Analysis and Design
Second Edition
These values apply for the situations in the table. We have used values of ry/y that are conservative for each of the
particular ranges of section type. If your structure doesn’t conform to one of the ones listed then you need to derive
a value of x0 for your particular structure using the equation for x0. Don’t forget that ry/y can never be greater than 1
for a symmetrical section. The values of x0 that are given by the equation and shown in the table are derived for
flexural buckling, that is buckled modes that are controlled by bending of some element or elements. When the
buckling is controlled by axial extension of, for instance, a bracing system, or by shear deflections, then there is no
direct method of calculating a correlation to BS 5950-Part 1:2000. In these ‘shear buckling’ situations it seems
prudent to use a value of x0 = Ld/250.
Mind you, don’t get too hung up on the detail of these calculations; BS 5950-Part 1:2000 says that we can ignore
these effects if they contribute less than 10% of the moment in a section. By implication it is pointless to try to
evaluate these effects to very high degree of accuracy. Also, the odds of the steel fabricator building the structure
with an initial imperfection that exactly resembles one of the buckled mode shapes that GSA has calculated are
pretty small.
We still need to define the length of the waveform Ld and the deflection assumed by the solution code for the
eigenvector at the middle of the waveform xg usually 1.0 for MAX normalization. In the case of the Euler strut, in
mode 1, the length of the buckled waveform is equal to the length of the strut. In mode 2, L is equal to half the
length of the strut, and in mode 3, L is equal to one third of the length of the strut.
54
Structural Engineering Analysis and Design
Second Edition
55
Structural Engineering Analysis and Design
Second Edition
For real structures, when we look at a real buckling mode the first thing that we need to establish is which member,
or members, is it that is a) causing the buckling and b) being affected by the buckling. When we say that a member,
or members, is ‘causing the buckling’ what we mean is that we are assuming that the structure has been built with
an imperfection in that member, or members (that mimics the shape of the buckled mode shape). In the case of the
simple strut in example 1 this means that the strut is initially bowed. In the case of the bracing in example 14 this
means that the bracing members were too long. When we say that a member is ‘affected by’ the buckling we mean
that moments or forces are induced in them, under the normal structural loads, as a result of the imperfection that
the structure was built with. Luckily, our definition of a buckling mode leads to the situation where the two sets of
members (a) and (b) above are generally coincident. A member that the analysis code identifies as having a high
curvature (or extension) becomes the one that has a curvature (or extension) as its initial imperfection. And it is no
coincidence that a member with high curvature (or extension) in the buckled mode is a member that has a high
moment (or axial force) in the force results for that mode. Let us call these members the primary buckling members
for the mode. Spotting the primary buckling members for any given mode is made easier by looking for the
maximum ‘total strain energy density’ in the structure in that mode. The total strain energy density is a value that
the analysis code calculates, and will display, either as a diagram, or as a contoured value. It is, roughly speaking,
defined as moment x curvature (or axial force x axial extension) divided by member area to give an energy density
per meter along the beam. Since the total strain energy density is the product of the two variables that we are most
interested in when we want to find the sets of members (a) and (b) above, it gives us a powerful graphical tool for
highlighting which the primary buckling members are.
56
Structural Engineering Analysis and Design
Second Edition
The resultant forces from each buckled mode can then be combined with the other ultimate forces in the structure
(from the P-∆ analyses) to give sets of design forces for the final stress check. Note that each set of resultant
buckling forces occurs because the structure was (we assume) built with a specific initial imperfection shape
that corresponded to the buckled mode shape. Therefore we do not need to combine forces from one mode
with forces from another since the structure cannot be built into both shapes at the same time.
The sign of q0i should be taken so as to maximize the total response in terms of section utilization. Usually
imperfections in the lowest buckling modes (λcr < 10) will cause the greatest additional response. Several such
modes may need to be considered, each maximizing the utilization of particular sections.
It is sometimes necessary to consider the effect of combining modal imperfections. The following plots show two
mode shapes of the Millennium Wheel. The rim buckles under the compression induced by the prestress in the
spokes. The load factors for the two modes are virtually identical, as are the mode shapes, except they are rotated
such that the position of nodes and antinodes are swapped.
It would be incorrect only to consider imperfections in each mode separately, as this would suggest that the
strength requirement for the rim varies around the circumference. A more general imperfection of the required
magnitude can be obtained by combining the imperfections due to sinα.q02 and cosα.q03. The same amplifier
applies to this combined imperfection, resulting in the same maximum response, except that the maximum can
occur anywhere around the rim. This supports intuition, the rim strength should be constant around the
circumference. Another such example is a long beam on elastic foundations, which has a numerous modes of very
similar shape and load factor. Again, an analysis using a generalized imperfection, based on combined modal
imperfections, would demonstrate that, away from the ends, the beam strength needs to be constant. Such paired or
multiple modes arise because structure and loading have a high degree of symmetry. The Millennium Wheel has
cyclic symmetry and the beam on elastic foundations has longitudinal repetition.
3.1.4.5 Design
It is quite normal to analyze structures assuming they remain linear elastic but then to check the member forces
against plastic capacities. Provided the system is not prone to buckling, partial plasticity causes redistribution of the
forces within the structure but it does not change the overall loading. To be strictly conservative when using linear
buckling analysis, the member forces would need to be limited to the elastic capacities, reduced to allow for the
effects of residual stresses. Codes of practice may allow for this by making the magnitude of the strut imperfection
depend on whether the member is to be checked elastically or plastically. A similar allowance can be made in the
proposed method by increasing the magnitude of the design imperfections.
57
Structural Engineering Analysis and Design
Second Edition
The simplified overall buckling check in BS 5950-Part 1:2000 Clause 4.8.3.3 does not amplify the applied
moments. While this is not conservative, it is usually argued that the effect is small. The more exact approach does
include moment amplification, as does the proposed method.
3.1.4.6 Limitations
The proposed first order analysis method will not be appropriate if the shape of the structure changes significantly
under the applied loading. The classic example would be very shallow arches that flatten under load and are prone
to snap through buckling. The axial force in the arch would be significantly underestimated if calculated on the
basis of the initial geometry, it needs to be calculated based on the reduced curvature of the flattened, loaded arch.
The significance of the effect could be assessed by considering the change in geometry arising from limiting elastic
strain in the arch.
58
Structural Engineering Analysis and Design
Second Edition
3.1.4.7 Methodology
Thus, to perform a linearized elastic buckling analysis, the following procedure is undertaken for each ultimate
state load combination.
(a) P-∆: Apply ultimate limit state load case combination onto the structure (hence the load factors will be
factors upon the ultimate limit state effects). Perform a P-∆ analysis; note two passes required.
(b) Eigenvalue: Perform the linear eigenvalue buckling analysis (two passes required) to obtain the lowest
buckling factor. If λcr1 > 10, then a linear static analysis (no P-∆ effects) may be sufficient.
(c) Modify load case combination to maximize the amplification terms (by minimizing λcr i.e. by
maximizing the geometric stiffness): Modify the load case combination in order to obtain an even
lower buckling factor by maximizing the geometric stiffness matrix. From observation of the critical
buckling mode shape, the loading that contributes to the geometric stiffness can be identified. This
should be maximized to increase the geometric stiffness, and hence obtain the lowest buckling load
factor. For example, for an arch this would mean identifying the load case combination causing the
maximum arch compression and for a column, the maximum axial compression. Again, of course, the
two passes (linear static on KEA and eigenvalue on KEA and KGA) must be performed for the eigenvalue
analysis. Hence the lowest (λcri < 10) buckling load factors and associated mode shapes are obtained.
Amplification terms
(d) Modify load case combination to maximize linear response of critical mode: Modify the load case
combination seeking to maximize the product of the linear response, dependant on the modal force
{φ1}T{f}.
(e) Iteration: Repeat steps 1-4 until the smallest buckling load factors and largest P-∆ response obtained.
(f) Incorporate modal imperfection responses: Calculate the modal imperfection responses as
x0 ry a Ld
q 0i = where x 0 = Ld =
xg y 1000 some constant
Ld
∴ q 0i =
x g . some constant
for every significant mode, i.e. all modes that give rise to significant forces in the structure. This
usually means the modes with λcr of less than 10.
(g) Total physical response: The total physical response will be the summation of the P-∆ case and the
modal imperfection responses (of the significant modes with λcr < 10) in physical coordinates.
Total response = P-∆ response +/- 1/(λcr1-1)q01 x {φ}1
Total response = P-∆ response +/- 1/(λcr2-1)q02 x {φ}2
…
Total response = P-∆ response +/- 1/(λcri-1)q0i x {φ}i
…
… etc … until λcr ~ 10
We do not need to combine forces from one mode with forces from another since the structure cannot
be built into both shapes at the same time. Envelope the results, hence incorporating the imperfections
in all the modes up to λcr ~ 10.
59
Structural Engineering Analysis and Design
Second Edition
(h) Design: Perform local linear elastic (not ULS plastic) capacity checks on the enveloped response. This
then becomes analogous to the code based Perry-Robertson element buckling check applied to global
modes.
Let us look at a real example. The Core Terminal Building roof at Terminal 5, Heathrow, London is not only a very
fine piece of structural engineering and a good example to look at. The model includes 2 rafters and a single pair of
support structures. This frame spans the width of the terminal building and will be repeated 11 times to cover the
length.
The first overall buckling mode looks, at first sight, to be major axis buckling of the rafters (which are prestressed
into compression as well as having an arched shape). It looks as though one goes up and the other goes down, with
each behaving rather like the strut in example 1. However, examination of the strain energy density diagram reveals
that it is the bracing in the roof plane that is the primary buckling member set, rather like example 13. The y
direction sway and the major axis moments in the rafters are secondary effects. The value of Ld has therefore been
taken as the distance between the centres of rotation of the two ends of the roof plane horizontal truss (as observed
from the deflected shape diagram when viewed from above). And xg has been taken as the horizontal deflection of
the centre of the truss relative to its two ends (as defined by those centres of rotation).
The second overall mode could also have been deceptive. The primary buckling members are clearly the main
rafters but what is causing the buckling? One might imagine that the structure was swaying in the east west
direction because of the heavy vertical loads on the legs, as in examples 9 or 10. However a simplified model
showed that the λcrit for that behaviour should be in the region of 40 rather than the value of 6 that the full model
predicts. It turns out that a simple model based on the action displayed in example 11 is a much closer match to the
behaviour of the full model. The rafter itself behaves rather like the strut in example 2, or indeed the beam in
example 11. Therefore the appropriate value of Ld is measured from the end, to the point of contra flexure in the
centre. And xg is taken as the vertical deflection of the rafter ¼ point (this is to avoid an overestimate of xg that
might result if the east west movement were included).
The third mode is rather like the first in that the primary buckling members are in the roof bracing. In this case the
action is a little more complex structurally. The centre bay of bracing diagonals has been omitted for visual reasons.
60
Structural Engineering Analysis and Design
Second Edition
This means that imperfections in the length of adjacent bracing diagonals have a significant effect, and shear of this
central bay forms the third mode. Again Ld is taken between points of rotation of the roof plane truss and xg has
been measured horizontally.
The fourth mode illustrates the need to keep our eyes open. The strain energy density diagram would lead us to
think that the primary buckling members were the roof plane bracing, like in modes 1 and 3. But the deflected
shape, and the bending moment diagram do not fit with this hypothesis. A simplified model confirmed that a
straight rafter in compression would buckle in a very similar shape at a similar λcrit. Therefore we conclude that the
mode is very like example 3. Ld is the length of the centre portion of rafter between the points of contra flexure. xg
is measured vertically, relative to the vertical deflection of the points of contra flexure.
In addition to these overall modes there are plenty of modes that show buckling of only one part of the structure. In
the case shown below it is the legs that are buckling. If you wanted to, you could use results from this analysis to
check the capacity of the leg CHSs themselves. It would be just as valid, however, and probably much quicker, to
use the code of practice for these sections. In this case Ld and xg would be just what you would expect from looking
at example 1, and so are the forces.
61
Structural Engineering Analysis and Design
Second Edition
SOL 101
$ CASE CONTROL SECTION
The applied loads are reported at the grid points, and include implicitly defined loads such as pressure or gravity.
62
Structural Engineering Analysis and Design
Second Edition
FORCE or MOMENT is used to apply a concentrated force or moment at a grid in either the basic or a user-
defined coordinate system.
$ BULK DATA
Grid Scale
FORCE SID CID N1 N2 N3
ID Factor
Grid Scale
MOMENT SID CID N1 N2 N3
ID Factor
To apply a concentrated force or moment at a grid point in or about a vector defined by two arbitrary grid points,
FORCE1 or MOMENT1 is used. To apply a concentrated force at a grid point in or about a vector defined by the
cross product of two vectors defined by four grid points, FORCE2 or MOMENT2 is used.
To define a linearly varying element load on CBAR, CBEAM and CELBOW elements, the PLOAD1 card is used.
$ BULK DATA
To apply a uniform normal pressure load on CQUAD4, CTRIA3 or CSHEAR surfaces defined by three or four
grids on one or many shell elements, the PLOAD card is used. The direction of the pressure is based on the right
hand rule and the ordering sequence of the grids.
$ BULK DATA
To apply uniform normal pressure load on CQUAD4, CTRIA3 or CSHEAR using element ids, the PLOAD2 card
is used. The direction of the pressure is based on the right hand rule and the element grid definition order.
$ BULK DATA
To apply a linearly varying not necessarily normal pressure load (such as hydrostatic pressure) on a surface of
any two or three-dimensional elements, the PLOAD4 card is used.
Note that point loads and moments on 1D elements are independent of the geometry of the element. Pressure loads
(applied in stress units) on 2D and 3D elements are dependent upon the dimensions of the loaded face of the
element. Line edge loads (applied in stress units) on 2D elements are dependent upon the thickness and the
dimension of the loaded edge.
63
Structural Engineering Analysis and Design
Second Edition
To apply acceleration forces on elements (such as force due to gravitational or centripetal acceleration), the GRAV
card is used. The GRAV card is used to simulate a load resulting from the inertia of the component of system.
Gravity may be one example. Another example may be the acceleration induced in a car component when the car
accelerates. The inertia of the component will resist the acceleration and hence produce a force in the same way as
it resists gravitational acceleration. Likewise, a centrifugal acceleration load allows rotating systems to be evaluated
statically.
$ BULK DATA
Scale
GRAV SID CID N1 N2 N3 MB
Factor
The force due to acceleration is calculated by multiplying the mass matrix with the acceleration specified. To
specify a gravitational acceleration of 9.81 m/s2, with the scale factor equaling 1.0, N3 = −9.81, with the vertical
direction being the global Z direction. The mass definition is defined by the density on a material entry.
The GRAV card cannot have the same ID as other static load types. Hence, a load combination must be specified
using the LOAD bulk data entry with the ID equal to that specified by the LOAD card in the Case Control Section.
The LOAD bulk data entry can be used to combine FORCEi, MOMENTi, PLOADi, GRAV, and SPCD cards.
$ BULK DATA
Overall
LOAD SID Scale S1 L1 S2 L2 S3 L3
Factor
S4 L4 …etc…
64
Structural Engineering Analysis and Design
Second Edition
There are two methods available for specifying an enforced displacement at a component. The first method is to
enter the value of the enforced displacement directly on an SPC entry. The alternate method to enforce a
displacement at a component is to use the SPCD Bulk Data entry. The SPCD entry is actually a force, not a
constraint, but it is used in conjunction with an SPC entry to enforce the displacement. When you use an SPCD
entry, internal forces are computed that are applied to the structure to produce the desired enforced displacements.
To specify a static enforced displacement at grid points, the SPCD card may be used.
$ BULK DATA
SPCD SID G1 C1 D1 G2 C2 D2
Grid points with an enforced displacement using the SPCD entry must also appear on an SPC or an SPC1 Bulk
Data entry. The SPCD method of enforcing a nonzero constraint is more efficient than using an SPC entry alone
when you are using multiple subcases that specify different constraint conditions. Note also that when you use an
SPCD entry, the displacement values entered on the SPC entry are ignored-only the SPCD values are used.
3.1.5.1.3 Lack-of-Fit
To specify a lack-of-fit or vice versa on line elements, the DEFORM card can be used.
$ BULK DATA
A positive deformation D1 refers to elongation and a negative D1 refers to a contraction (lack-of-fit). When using
the DEFORM entry, you must remember that in general you are not enforcing a strain or an actual extensional
length to the element. What you are doing is applying a force to the element that produces the specified extension if
the element is free to expand without internal forces being generated. This computed force is added to the other
forces in the model. Since most elements in your model are not free to expand, the extension value you specify may
not be achieved.
The prestress is an element (such as a prestressed cable) can be modeled using temperature loads. The axial force
generated from applying the temperature load on a line element that is totally fixed at its nodes can be calculated as
follows,
faxial = EAδ/L
faxial = EAεstrainL/L
faxial = EAεstrain
where εstrain = α ∆T, α is the coefficient of thermal expansion (per °C) and ∆T is the change in temperature in °C.
Note that the loads induced by the temperature are independent of the length of the element, L.
However, it must be realized that the force calculated above is based on a fixed end line element. In reality and in
the numerical model, the ends will in general not be fixed. Hence the element will contract with tensile temperature
loading causing the force to decrease. Thus the specified ∆T will have to correspond to a force which is greater
than the intended final prestress. Hence an iterative procedure is required (hence the need to perform the analysis
repeatedly) until the required prestress is achieved in the final deflected configuration.
65
Structural Engineering Analysis and Design
Second Edition
Thermal expansion and contraction is a structural rather than a thermal solution i.e. it is the analysis of stress and
displacement due to thermal effects that is being considered, not the temperature at a point in the structure, which
requires a heat transfer analysis. In fact, a heat transfer analysis can actually be used to generate the temperature
profile and distribution and these results can in turn be applied as temperature loads in a thermal stress analysis.
The input data required for analyzing thermal expansion is the coefficient of thermal expansion and the temperature
distribution in model. The temperature data and the thermal expansion coefficients (in the material entry cards) are
used internally to calculate equivalent forces and moments acting at grid points. Temperatures can be specified at
grid point using TEMP cards, or at elements using TEMPRB (line elements) or TEMPP1 (shell elements).
Average temperature specified directly for an element will take precedence over the temperature interpolated from
the elements connected grid points. Solid elements get their temperature only by interpolation from connected grid
points. Temperature load is calculated for a fixed ended member as
faxial = EAεstrain
where εstrain = α ∆T, α is the coefficient of thermal expansion (per °C) and ∆T is the change in temperature in °C
This equivalent nodal load is only attained if the element is constrained from expanding and contracting. Note that
the loads induced by the temperature are independent of the length of the element, L. However, if the elements is
not constrained from expanding and contracting, then there will be a lateral deflection of faxial/k = faxial/(EA/L) =
EAεstrain/(EA/L) = εstrainL which clearly is dependent upon the length of the element, L. Temperature loads are
defined as follows.
TEMPD is to define temperatures at all grid points not given a temperature. It is good practice to define these and
the initial nodal temperatures. TA and TB are the variations of temperature ∆T from the reference temperature
defined by TEMP (INIT).
Note also that SUBCOM cannot be used conventionally to combine temperature load cases. The following
procedure is required. SUBCOM 3 combines the mechanical loads of load cases 1 and 2, but not their temperature
loads. Temperature loads must thus be specified separately calling a separate TEMPi bulk data entry card, which
can be made to combine the temperature loads of the load cases 1 and 2.
66
Structural Engineering Analysis and Design
Second Edition
3.1.5.1.6 Linear Contact Definition with MPC Equations for SOL 101
Contact definitions (normal to contacting surfaces, no sliding friction) can be made within a linear static analysis
(albeit using an iterative method). RBE2 elements can be defined between adjacent grids of surfaces, then a linear
static analysis performed, then the RBE2s in tension deleted (as a contact cannot sustain tension), further linear
static analyses performed and further RBE2 elements deleted until there are no RBE2 elements in tension within
the contact interface. Although this method works, it is extremely cumbersome, as every load case will have a
different contact distribution. This process can be automated far more elegantly using MPC equations in SOL 101.
It uses a iterative constraint type approach. The constraints are applied to grid points or SPOINTS. The constraint
ensures that:
1. The displacement (UR) cannot be negative. This is to ensure that there is no penetration. Therefore, the
chosen degree-of-freedom must be perpendicular to the contact surface and positive in the opening direction.
2. The force of constraint (QR) cannot be negative. This is to ensure that there is no tension.
The constraints are satisfied by an iterative technique that is built into SOL 101. The iterative process starts with a
random vector. This random vector assumes certain grids to be in contact and other grids to be in an open state. A
solution is obtained when all the gap constraints are satisfied, i.e., there's no penetration and no tension forces. If a
limit cycle (return to a previous state) is detected during the iterations, a new random start vector is tried. This
approach provides an alternate method to the use of GAP elements in SOL 106. Some experiments have shown that
the cost of analysis will be about the same. The advantage is that you do not have to learn how to calculate the GAP
stiffness nor how to control SOL 106. Multiple load conditions are allowed, and each will be solved separately. If
the constraint is between a finite element model and a fixed boundary, then arrange to have one of the degrees-of-
freedom at the boundary grid points represent motion perpendicular to the boundary. A positive displacement
represents motion away from the boundary. If, on the other hand, the constraint represents relative motion between
two bodies, MPC equations are needed to define a relative motion degree-of-freedom, which is then constrained to
have a non-negative displacement. Consider the following contact interface between two faces of the FE model
UG1 UG2
Positive U Direction
S0
The MPC for each and every pair of grids is defined as
S = UG2 – UG1 + S0
where S is the relative motion DOF, UG2 – UG1 arranged such that a positive value indicates opening of the contact
and S0 representing the initial gap opening, zero if the model interface gap is zero. The MPC is rearranged
S – UG2 + UG1 – S0 = 0 To model a fixed contact interface between a surface of the FE model and
for the definition of the card. a fixed boundary, simply omit U to represent a fixed boundary.
G1
1 S 0 G2 U
MPC (Relative Motion (SPOINT 1.0 (G2 -1.0
(MPC ID) (G2 ID)
SPOINT ID) Component) Component)
G1 U S0 0
(G1 ID)
(G1 1.0 (Initial Gap (SPOINT -1.0
Component) SPOINT ID) Component)
S 0
(Relative
SUPORT Motion
(SPOINT
Component)
SPOINT ID)
67
Structural Engineering Analysis and Design
Second Edition
S
(Relative
S0
SPOINT (Initial Gap
Motion
SPOINT ID) SPOINT ID)
PARAM, CDPRT requests the printing of constraint violations (in terms of UR for negative displacements
denoting overlap and QR denoting negative forces of constraints) during the iterations. The output is as follows.
Finally, the intention would be to find out the value of the constraint forces. This is best found from MPCFORCES
(PRINT) from the G TYPE Grid output. S TYPE SPOINT forces also correspond.
The same values are also presented in the SPCFORCES output (only for SPOINTS), somewhat redundantly.
68
Structural Engineering Analysis and Design
Second Edition
3.1.5.1.7 Modelling Centrifugal Force in SOL 101– Modelling Steady State Static Equilibrium Conditions
Centrifugal force is applied onto systems assumed to be in a steady-state, equivalent to static equilibrium. For
instance, the forces acting on a car spinning in a test track at a certain velocity and/or acceleration can be
ascertained using linear static analysis with centrifugal force as the system is in static equilibrium. Likewise the
forces in components rotating in a rotating drum can be ascertained. The six rigid-body boundary conditions must
of course still be applied onto the static system. The RFORCE entry is used to apply the force due to rotational
velocity and/or acceleration. On the RFORCE entry, the components of a spin vector used internally to compute
centrifugal forces are input. Each component of the spin vector is multiplied by the same scale factor. For instance
consider the two rotating masses shown below. The masses are accelerated at constant angular acceleration of 20
rev/sec2 from 0 to 200 rev/min. The goal is to determine the CBAR axial force as a function of angular velocity.
SPC
From graph, note that centrifugal forces are not linear with angular velocity, but proportional to ω2. Hence, using
a SUBCOM command to scale the subcases with RFORCE entries can lead to misleading results if abused.
Another interesting note is that the model really does not rotate; it is actually fixed at the center. If the constraints
permitted the model to rotate, the run would fail. RFORCE is called by the LOAD case control command.
$ BULK DATA
Velocity Scale
Grid
RFORCE SID CID Factor A = R1 R2 R3 METHOD
ID
...rev/s
Acceleration
Scale Factor
RACC = 20rev/s2
69
Structural Engineering Analysis and Design
Second Edition
Method=1 yields correct results only when there is no coupling in the mass matrix. This occurs when the lumped
mass option is used with or without the ZOFFS option (see the CQUAD4 entry for a description of ZOFFS).
Method=2 yields correct results for lumped or consistent mass matrix only if the ZOFFS option is not used. The
acceleration terms due to the mass offset (X1, X2, X3) on the CONM2 entry are not computed with method=2. The
possible combinations are
In addition, for problems with elements that have edge grid points (CQUAD8, CTRIA6, CTRlAX6, CHEXA,
CPENTA, and CTETRA), correct centrifugal loads are produced only if the parameter PARAM, COUPMASS, x
(where x is greater than 1), is included and Method 2 is used.
70
Structural Engineering Analysis and Design
Second Edition
In order to perform a P-∆ static analysis, the geometric stiffness matrix must be calculated. However, because the
geometric stiffness matrix is dependent upon the state of element stresses (or the forces in the elements), it is
necessary to perform a static analysis without the geometric stiffness as a first pass, and then only perform another
static analyses incorporating the geometric stiffness matrix. The following procedure can be adopted.
Phase 1
Perform static analysis (with loads that cause the greatest negative or positive geometric stiffness) based on [KEA]
but include the segyroa.v2001 alter prior to the Case Control Section in order to compute the geometric stiffness
matrix [KGA] (and output into a .pch file) based on the generated element loads from the [KEA] static analysis.
Phase 2
Perform static analysis with a general loading function based on [KEA] + [KGA] by including the k2gg = ktjj
statement in the Case Control Section and the outputted .pch file which contains the ktjj matrix in the Bulk Data.
The responses at this stage represent the P-∆ (KGA From KEA) static response to the general loading function.
The method is valid when only the prestress is judged to affect the geometric stiffness such as in the
compressive preload of building columns due to gravitational loads and the prestressing of extremely taut cables
that sag very little under gravity but not in systems such as suspension bridges. Where lateral loads are large
enough to affect the geometry of the system with prestress, then a repetitive P-∆, SOL 106, implicit dynamic
relaxation SOL 129 or explicit dynamic relaxation must be employed. But in single P-∆ analysis, because cables do
not have much elastic bending stiffness, the initial static preload subcase should only include the prestress and not
gravity as including gravity is the same as solving two linear static problems of stiffness KEA with preload and
gravity as the applied loads respectively. Clearly, in the gravity case, it is nonsensical as the cables do in reality
have differential stiffness (from the prestress) to resist the gravitational force. Prestress in one direction (i.e. along
the axis of cable) will cause a differential stiffness in the orthogonal direction. Gravity acts in the orthogonal
direction and hence cannot be accounted for in the calculation of the prestress in this single P-∆ analysis. To
quantitatively decide if gravity need not be considered in contributing to the differential stiffness of the cables, a
static P-∆ analysis should be carried out, the first subcase being a SOL 101 with only the prestress as applied loads
and the second subcase a P-∆ SOL 101 (i.e. utilizing the induced prestress from the first subcase to form a
geometric stiffness matrix) with both the gravity and prestress included as applied loads. If the difference in the
cable element forces between subcases 1 and 2 is negligible, then gravity has little influence in affecting the
geometric stiffness. If there is a major difference in the cable element force, then clearly, gravity will affect the
geometric stiffness and as such, a repetitive P-∆ , SOL 106, implicit dynamic relaxation or explicit dynamic
relaxation must be used to converge to the true KT. Likewise, in the single P-∆ analysis of multi-storey buildings,
gravity (and only gravity) acts in axis of columns to generate prestress, and the differential stiffness is computed for
the orthogonal direction reducing resistance to lateral wind forces, applied in the second subcase with gravity too.
The segyroa.v2001 alter computes the differential stiffness due to the prestress and also the follower force. The
follower force is calculated and incorporated by the use of PARAM, FOLLOWK, YES. We know how the prestress
affects the differential stiffness, namely a tensile prestress causing an increase in stiffness. The effect of the follower
force on the stiffness is different. For example, for a cylinder under external pressure critical buckling load may be over-
estimated (even though the mode shapes are similar) in a SOL 105 and the natural frequencies in vibration may be under-
estimated (even though the mode shapes are similar) in a SOL 103 in the absence of follower stiffness. To the contrary,
this observations are reversed in case of centrifugal loads. Centrifugal forces as a constant (static) load are applied by a
Bulk Data RFORCE to any elements that have masses. The follower stiffness due to centrifugal load has the effect of
lowering stiffness (although the centrifugal load tensioning effect increases stiffness), consequently lowering natural
frequencies (even though the mode shapes are similar) in a SOL 103 and lowering the buckling loads (even though the
mode shapes are similar) in a SOL 105. This effect increases as the RPM increases, and it becomes significant when the
RPM is over 1000.
71
Structural Engineering Analysis and Design
Second Edition
3.1.5.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Static Analysis
To obtain KTA, to be theoretically exact, a GNL SOL 106 (or implicit dynamic relaxation SOL 129 or explicit
dynamic relaxation LS-DYNA) with prestress (as temperature loads say) and gravity must be undertaken.
Alternatively, an approximation to KTA can be obtained by repetitive P-∆ static analyses with the prestress (as
temperature loads say) and gravity applied. The procedure to obtain this approximate KTA will be presented. Note
that the approximate KTA will be the summation of the elastic stiffness KE at the undeflected (by the prestress and
gravity) state but KG at the deflected (by the prestress and gravity) state. Hence if KE changes considerably during
the application of the prestress, a full SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic
relaxation LS-DYNA), which converges to the KE and KG at the deflected (by the prestress and gravity) state
should be employed. Hence for the modelling of a suspension bridge where there is a great change in geometry
(known in the bridge industry as form-finding, so-called because it is necessary to find the form or shape of the
catenary suspension cables), it may be prudent to employ SOL 106 (or implicit dynamic relaxation SOL 129 or
explicit dynamic relaxation LS-DYNA), but for a high tension low sag cable on say a tower with prestressed
cables, the repetitive P-∆ static analysis may be adequate. The repetitive P-∆ analysis basically involves a number
of iterations of linear static analyses to obtain an approximate KTA. Note again that A refers to the initial
undeflected (by the collapsing load) state, but deflected by the prestress and gravity. To perform the repetitive P-∆
analysis, a static analysis is performed based on KEA with temperature loads and gravity to generate forces in the
structural elements, which in turn provides input for the computation of KGiAKTm where m is the iterations.
Repetitive static analysis is performed with the prestress and gravity updating the stiffness matrix KEA + KGiAKTm-1 +
KGiAKTm until convergence of displacements is obtained. The tangent stiffness at this stage is the approximate
converged tangent stiffness matrix KTA = KEA + KGiAKT. The converged displacements represent the approximate P-
∆ (KGA From Approximate KTA) static response to the initial prestress loads. The converged geometric stiffness at
this stage would be that based upon the approximate tangent stiffness matrix KTA, i.e. KGiAKT.
Phase 1
Perform static analysis (with prestress and gravity) based on [KEA] but include the segyroa.v2001 alter prior to the
Case Control Section in order to compute the geometric stiffness matrix [KGA]1 (and output into a .pch file) based
on the generated element loads from the [KEA] static analysis.
Phase 2
Perform static analysis (with prestress and gravity) based on [KEA] + [KGA]1 by including the k2gg = ktjj statement
in the Case Control Section, the outputted .pch file which contains the ktjj matrix in the Bulk Data and the
segyroa.v2001 alter prior to the Case Control Section to compute the [KGA]2 (and output into the .pch file
overwriting previous data) based on the generated element loads from the [KEA] + [KGA]1 static analysis.
Phase 3
Repeatedly perform the Phase 2 static analysis (with prestress and gravity) based on [KEA] + [KGA]i for i = 2 to n
where n represents the number of iterations required for the change in deflections between analyses to become
negligible. This would signify that the change in the [KGA] matrix become negligible and the correct [KGA] is
attained. The deflections and the other responses at this stage represent the P-∆ (KGA From Approximate KTA) static
response to the prestress and gravity. The stiffness of the structure is KTA.
Phase 4
The P-∆ response to a general loading function can now be ascertained by performing a P-∆ static analysis with the
converged tangent stiffness of the structure, KTA. The responses at this stage represent the P-∆ (KGA From
Approximate KTA) response to the general loading function.
72
Structural Engineering Analysis and Design
Second Edition
3.1.6.1 Static Displacements by the Unit Load Method of the Virtual Work Principle
The integrals can be evaluated using classical calculus integration methods, product integrals or using the
numerical Simpson’s Rule I = h/3[f0+4f1+f2]. Simpson’s Rule is exact for the multiplication of continuous functions
up to and including cubic variations. Non-continuous piecewise functions should be split at the discontinuities and
integrated in sections.
73
Structural Engineering Analysis and Design
Second Edition
74
Structural Engineering Analysis and Design
Second Edition
75
Structural Engineering Analysis and Design
Second Edition
3.1.6.2 BMD and SFD of Structures of Low Static Indeterminacies by Flexibility Analysis
76
Structural Engineering Analysis and Design
Second Edition
77
Structural Engineering Analysis and Design
Second Edition
78
Structural Engineering Analysis and Design
Second Edition
79
Structural Engineering Analysis and Design
Second Edition
80
Structural Engineering Analysis and Design
Second Edition
81
Structural Engineering Analysis and Design
Second Edition
82
Structural Engineering Analysis and Design
Second Edition
83
Structural Engineering Analysis and Design
Second Edition
84
Structural Engineering Analysis and Design
Second Edition
85
Structural Engineering Analysis and Design
Second Edition
86
Structural Engineering Analysis and Design
Second Edition
87
Structural Engineering Analysis and Design
Second Edition
88
Structural Engineering Analysis and Design
Second Edition
89
Structural Engineering Analysis and Design
Second Edition
90
Structural Engineering Analysis and Design
Second Edition
91
Structural Engineering Analysis and Design
Second Edition
92
Structural Engineering Analysis and Design
Second Edition
93
Structural Engineering Analysis and Design
Second Edition
94
Structural Engineering Analysis and Design
Second Edition
95
Structural Engineering Analysis and Design
Second Edition
96
Structural Engineering Analysis and Design
Second Edition
97
Structural Engineering Analysis and Design
Second Edition
3.1.6.3 Bending Moments, Shear Force, and Displacements of Structural Elements by the Stiffness Method
The flexural stiffness relations for a beam element gives the relationships between the actions shear force &
bending moment and the kinematics displacements and rotations.
The force-displacement relationship of the Timoshenko beam is as follows. For the Euler-Bernoulli beam, β is
zero.
F1 12 6L − 12 6L δ1
M (4 + β)L 2
− 6L (2 − β )L θ1
2
1 EI 6L 12EI
= 3 β=
F2 L (1 + β ) − 12 − 6L 12 − 6 L δ 2 GAL2
M 2 6L (2 − β) − 6L (4 + β )L2 θ 2
98
Structural Engineering Analysis and Design
Second Edition
3.1.6.4 Bending Moments and Shear Force of Structures of Low Kinematic Indeterminacies by Moment
Distribution
Fixed end forces due to distributed loading is −/+ωL2/8 and that due to a mid point load is −/+PL/8. The drop-down
due to distributed loading is ωL2/8 and that due to a point load is PL/4. The distribution factors summary is as
follows
The analysis of a structure with kinematic indeterminacy αk = 1 is exact, and those with αk > 1 is iterative. The
kinematic indeterminacies are reduced by
(a) changing moment/rotational stiffness of pinned end beam to 3EI/L from 4EL/L, hence eliminating the pinned
ended kinematic
(b) employing symmetry and anti-symmetry of structure and loading as follows
(i) axis of symmetry through a joint and symmetric loading
φcentre = 0, hence analyze half structure with equivalent central rotational fixity
(ii) axis of symmetry through a joint and anti-symmetric loading
BMcenter = 0, hence analyze half structure with equivalent central pinned end
(iii) axis of symmetry through a member and symmetric loading
analyze half structure with the stiffness of middle member of 2EI/L
(iv) axis of symmetry through a member and anti-symmetric loading
analyze half structure with the stiffness of middle member of 6EI/L
Hence, the analysis of pinned ended two-span continuous beams and propped cantilever is exact.
99
Structural Engineering Analysis and Design
Second Edition
100
Structural Engineering Analysis and Design
Second Edition
101
Structural Engineering Analysis and Design
Second Edition
102
Structural Engineering Analysis and Design
Second Edition
103
Structural Engineering Analysis and Design
Second Edition
104
Structural Engineering Analysis and Design
Second Edition
105
Structural Engineering Analysis and Design
Second Edition
106
Structural Engineering Analysis and Design
Second Edition
107
Structural Engineering Analysis and Design
Second Edition
108
Structural Engineering Analysis and Design
Second Edition
109
Structural Engineering Analysis and Design
Second Edition
110
Structural Engineering Analysis and Design
Second Edition
111
Structural Engineering Analysis and Design
Second Edition
112
Structural Engineering Analysis and Design
Second Edition
113
Structural Engineering Analysis and Design
Second Edition
114
Structural Engineering Analysis and Design
Second Edition
115
Structural Engineering Analysis and Design
Second Edition
116
Structural Engineering Analysis and Design
Second Edition
117
Structural Engineering Analysis and Design
Second Edition
118
Structural Engineering Analysis and Design
Second Edition
119
Structural Engineering Analysis and Design
Second Edition
120
Structural Engineering Analysis and Design
Second Edition
121
Structural Engineering Analysis and Design
Second Edition
122
Structural Engineering Analysis and Design
Second Edition
123
Structural Engineering Analysis and Design
Second Edition
124
Structural Engineering Analysis and Design
Second Edition
125
Structural Engineering Analysis and Design
Second Edition
126
Structural Engineering Analysis and Design
Second Edition
127
Structural Engineering Analysis and Design
Second Edition
128
Structural Engineering Analysis and Design
Second Edition
129
Structural Engineering Analysis and Design
Second Edition
130
Structural Engineering Analysis and Design
Second Edition
131
Structural Engineering Analysis and Design
Second Edition
132
Structural Engineering Analysis and Design
Second Edition
133
Structural Engineering Analysis and Design
Second Edition
134
Structural Engineering Analysis and Design
Second Edition
135
Structural Engineering Analysis and Design
Second Edition
136
Structural Engineering Analysis and Design
Second Edition
137
Structural Engineering Analysis and Design
Second Edition
138
Structural Engineering Analysis and Design
Second Edition
139
Structural Engineering Analysis and Design
Second Edition
140
Structural Engineering Analysis and Design
Second Edition
141
Structural Engineering Analysis and Design
Second Edition
142
Structural Engineering Analysis and Design
Second Edition
143
Structural Engineering Analysis and Design
Second Edition
144
Structural Engineering Analysis and Design
Second Edition
145
Structural Engineering Analysis and Design
Second Edition
146
Structural Engineering Analysis and Design
Second Edition
147
Structural Engineering Analysis and Design
Second Edition
148
Structural Engineering Analysis and Design
Second Edition
149
Structural Engineering Analysis and Design
Second Edition
150
Structural Engineering Analysis and Design
Second Edition
151
Structural Engineering Analysis and Design
Second Edition
152
Structural Engineering Analysis and Design
Second Edition
153
Structural Engineering Analysis and Design
Second Edition
154
Structural Engineering Analysis and Design
Second Edition
155
Structural Engineering Analysis and Design
Second Edition
156
Structural Engineering Analysis and Design
Second Edition
157
Structural Engineering Analysis and Design
Second Edition
158
Structural Engineering Analysis and Design
Second Edition
159
Structural Engineering Analysis and Design
Second Edition
160
Structural Engineering Analysis and Design
Second Edition
161
Structural Engineering Analysis and Design
Second Edition
162
Structural Engineering Analysis and Design
Second Edition
163
Structural Engineering Analysis and Design
Second Edition
3.1.6.6 Member Buckling Check Using The P-∆ Method To Incorporate Imperfections, Residual Stresses
3.1.6.6.1 ML, GL (P-∆ Based) Perry-Robertson Buckling of Column (Primary Stress is Only Axial)
Elements With Imperfections and Residual Stresses
Euler buckling considers only the axial stresses with no provisions for bending stresses. Hence, it is only valid for
perfectly straight columns. Perry-Robertson on the other hand includes the decrease in capacity due to initial
imperfections of the column. It is based upon finding the mean axial stress that will cause the onset of yielding of
extreme fibers due to the combined effects of both axial stresses and longitudinal bending stresses. Hence Perry-
Robertson implicitly includes yielding of the column at low slenderness ratios. It can be shown starting from the
basic equation for the onset of outer fiber yielding (utilizing the amplification factor derived in Section 3.1.2, hence
the reason it is called buckling by the P-∆ method)
PPR M 1 Px 0 1
+ = σy i.e. σ PR + = σy
A Z 1 − PPR Z 1 − PPR
PE PE
σ y + (1 + η )σ E σ y + (1 + η)σ E
2
Ax 0 x 0 y x 0 y
σ PR = − − σyσE where the Perry factor, η = = 2 =
2 2 Z r r r
The Perry factor η above is the theoretical imperfection factor which is dependent upon the initial eccentricity,
x0 and also the cross sectional properties. But BS 5950-Part 1:2000 goes a step further and requires an increase in
the Perry factor due to residual stresses from rolling and welding. It thus states that the Perry factor should be
a (L e − L o )
η= a = Robertson constant
1000 ry
The Robertson constant, a divided by 1000, i.e. a/1000 is equivalent to (x0/L)(y/r) and includes the initial
imperfection, the section shape and also residual stresses. The Lo plateau is introduced so that for a very low
slenderness, the strut reaches its full yield capacity. As an aside, note that although the radius of gyration of an area
has no physical meaning, we can consider it to be the distance (from the reference axis) at which the entire area
could be concentrated and still have the same moment of inertia as the original area.
Note that σPR approaches σE for large slenderness ratios i.e. for very slender columns; This means that the
sensitivity of slender columns to imperfections is less than that of stocky columns.
σ cr
σ cr = σ y
σy
σ cr = σ E
σ cr = σ PR
Le
r
Le π2E
=
r σy
164
Structural Engineering Analysis and Design
Second Edition
The capacity to buckling should be checked about both the major and minor axes. The slenderness is a function
of the cross section and the boundary conditions, hence the critical buckling capacity may occur about either the
major or minor axes. The critical load capacity is
PPR = σ PR A
Note that for welded I, H or box sections, pc should be obtained from Table 24 or calculated using a py value
lessened by 20 N/mm2. A fundamental parameter is the effective length, LE.
165
Structural Engineering Analysis and Design
Second Edition
Another method used to determine LE in buildings given in BS 5950-Part 1:2000 is based on the adjacent member
stiffnesses to determine coefficient k1 and k2 from which design charts are used to determine LE for sway and non-
sway frames.
Non-Sway Sway
166
Structural Engineering Analysis and Design
Second Edition
where
167
Structural Engineering Analysis and Design
Second Edition
168
Structural Engineering Analysis and Design
Second Edition
On top of the Perry-Robertson flexural buckling, columns are also subject to torsional buckling and flexural-
torsional buckling. These occur when the ratio of torsional to flexural stiffness is very much reduced.
169
Structural Engineering Analysis and Design
Second Edition
3.1.6.6.2 ML, GL (P-∆ Based) Secant Buckling of Column (Primary Stress is Axial) Elements With
Eccentric Loading, Initial Imperfections and Residual Stresses
The Secant buckling stress for an eccentrically (with eccentricity e) loaded column is as follows. It is based upon
finding the mean axial stress that will cause the onset of yielding of extreme fibers due to the combined effects of
both axial stresses and longitudinal bending stresses, the latter of which is primarily a result of the eccentricity of
the axial load. The initial imperfections and residual stresses can also be added to the loading eccentricity, e for an
equivalent eccentricity, e.
σy
σ SECANT =
ec L σ SECANT
1 + 2 sec e
r 2r E
The equation is an implicit formula, and hence iteration is required to solve for σSECANT. Note that the radius of
gyration, r is the minor axis radius of gyration. Also, c denotes the distance from the neutral axis to the extreme
fibre on the concave or compression side of the bent column. The term ec/r2 is called the eccentric ratio, a value
which is often calibrated to physical tests. In fact, it is the same as the Perry factor, η had there been no eccentricity
of the loading. Typical conventional design of structural steel columns employs a value of 0.25 for ec/r2.
170
Structural Engineering Analysis and Design
Second Edition
3.1.6.6.3 ML, GL (P-∆ Based) Lateral-Torsional Buckling of Beam (Primary Stress is Only Bending)
Elements With Imperfections and Residual Stresses
A critical parameter is the effective length, LE and is defined in clauses cl. 4.3.5.1 to cl. 4.3.5.5 for
Note that LLT is the segment length between restraints, and the span length is L. Note that the destabilizing loading
condition should be taken where a load is applied to the top flange of a beam or a cantilever, and both the load and
the flange are free to deflect laterally (and possibly rotationally also) relative to the centroid of the cross-section.
cl. 4.3.6.5
Bending strength, pb = f(py,λLT)
Table 16 or Table 17
cl. 4.3.6.6
Equivalent uniform moment factor, mLT
Table 18
Ensure LTB moment capacity Mb/mLT > applied design moment Md
Lateral torsional buckling cannot occur in beams loaded in their weaker principal plane; under the action of
increasing load they will collapse simply by plastic action and excessive in-plane deformation. Hence, the check
need only be made for bending about the major axis.
Lateral torsional buckling cannot occur for sections which are double symmetric such as a SHS, this being
accounted for by the fact that λLT becomes zero when Ixx = Iyy.
The equivalent uniform moment factor, mLT is used when the bending moment is not uniform across the length of
the beam. A value of 1.0 is obviously conservative. For the destabilizing loading condition mLT should be taken as
1.0.
171
Structural Engineering Analysis and Design
Second Edition
where
and where the Robertson constant aLT should be taken as 7.0 and the limiting equivalent slenderness, λL0 is
Restraints may be deemed to provide adequate stiffness and strength if capable of resisting a lateral force of at
least 2.5% of the maximum factored force in the compression flange.
172
Structural Engineering Analysis and Design
Second Edition
173
Structural Engineering Analysis and Design
Second Edition
174
Structural Engineering Analysis and Design
Second Edition
3.1.6.6.4 ML, GL (P-∆ Based) Buckling of Beam-Column Elements (Primary Stress is Axial and Bending)
With Imperfections and Residual Stresses
For Perry-Robertson flexural buckling about the major or minor axes with additional primary moments about
the major and minor axes, the following expression must be satisfied
Although unconservatively no amplification of major axis moment
considered, conservatively the minor axis elastic capacity is used.
Note that Pc is the Perry-Robertson flexural buckling capacity and is the smaller of Pcx and Pcy. The major axis
flexural buckling factor, mx and the minor axis flexural buckling factor, my are obtained from Table 26 (cl.
4.8.3.3.4).
For lateral torsional buckling with Perry-Robertson flexural buckling about minor axis and additional
primary moment about minor axis, the following expression must also be satisfied.
Although unconservatively no amplification of minor axis moment
considered, conservatively the minor axis elastic capacity is used.
Note that Pcy is the minor axis Perry-Robertson flexural buckling capacity. MLT is the applied major axis moment.
My is the applied minor axis moment. Mb / mLT is the lateral torsional buckling capacity.
More Exact Approach for I- or H- Sections with Equal Flanges (cl. 4.8.3.3.2)
For Perry-Robertson flexural buckling about the major with additional primary moments about the major or
minor axes, the following expression must be satisfied
where myx is obtained from Table 26 (cl. 4.8.3.3.4). The major axis flexural buckling factor, mx is also obtained
from Table 26 (cl. 4.8.3.3.4). Note that Pcx is the major axis Perry-Robertson flexural buckling capacity. Mx and My
are the major and minor axes applied moments. Mcx and Mcy are the major and minor axes plastic moment
capacities.
For lateral torsional buckling (of course about minor axis) with Perry-Robertson flexural buckling about
minor axis and additional primary moment about minor axis, the following expression must also be satisfied.
Similar more exact equations exist for CHS, RHS and box sections with equal flanges in cl. 4.8.3.3.3.
175
Structural Engineering Analysis and Design
Second Edition
176
Structural Engineering Analysis and Design
Second Edition
3.1.6.7 Overall Portal Frame and Multi-Storey Building P-∆ Analysis Using The Amplified Sway Method
Sufficient lateral stability must be provided to prevent instability. Sufficient strength must be provided to take the
increased forces (P-∆ effects) as a result of the slenderness.
To investigate the susceptibility of a structure to P-∆ effects, the elastic buckling load factor is calculated. Methods
based on BS 5950-Part 1:2000 for calculating λcr is given in Section 3.2.5.2.2.
Then assuming 4 < λcr <10, the P-∆ effects should be taken into account for the real horizontal loads (wind)
according to the amplified sway method. Had λcr been greater than 10, then the building is considered non-sway
and amplification effects, m on lateral forces need not be considered. For portal frames, if under the Notional
Horizontal Forces (NHF) of the gravity load combination, the horizontal deflections at the tops of the column are
less than height/1000, P-∆ effects need not be considered. Had λcr been less than 4, then a full nonlinear static
analysis must be undertaken to account for the P-∆ effect.
The amplified sway factor for the real horizontal loads (wind) (in perfect accordance with theory presented in
Section 3.1.2) is (cl. 2.4.2.7)
λ cr
m=
λ cr − 1
For clad structures with the stiffening effect of masonry infill wall panels or diaphragms of profiled steel sheeting
the amplified sway factor is decreased to
λ cr
m= but m > 1.0
1.15λ cr − 1.5
The ULS load combinations are presented in Section 8.2.3. Note the load case corresponding to the NHF includes
the factored dead and live load.
177
Structural Engineering Analysis and Design
Second Edition
3.2.1 Linearization of Tangent Stiffness Matrix and Formulating The Linear Eigenvalue Problem
λ
ML, GL Static Analysis
ML, GL Buckling Analysis λECR
ML, GL P-∆ Static Analysis
This section describes a simple and practical method of allowing for the effects of overall buckling of the structure.
While the method can be applied to the buckling of single members, this is more easily dealt with using
conventional codified methods. Where conventional codified methods falls short, is when the boundary conditions
of the structural members (which determines the all important buckling effective length) is not intuitive.
Linear elastic buckling analysis is the prediction of the critical loads (and their associated buckling mode shapes)
when the structure is at the point of elastic instability. This instability occurs when the tangent stiffness matrix
becomes singular as a result of the increase in the negative stiffness contribution of the geometric (differential)
stiffness.
At collapse, the structure has buckled (experienced elastic instability) but has no fully plastic hinges i.e. no
considerations of material nonlinearity have been taken into account. Hence this method of analysis is only valid if
firstly, the structure remains materially linear elastic throughout until the application of the critical load causes an
elastic instability. Secondly, it is a geometrically linear method of analysis. Hence the deflections must be small
enough for the stiffnesses to be based upon the initial undeflected (by the collapsing load factor) geometry.
Let us denote the initial undeflected (by the collapsing load factor) configuration by A. In the (KGA From KEA)
elastic buckling analysis, at collapse, the elastic stiffness matrix at the initial undeflected (by the collapsing load
factor) configuration is balanced by the geometric (differential) stiffness due to the load factor.
[[K E ]+
A
]
λ[ K GAKE ] {φ} = {0}
AKE
KG represents the geometric stiffness matrix which was calculated based on the small displacements obtained by
solving the system (with the collapsing load) with stiffness KEA. Hence, a static analysis is first performed (a first
pass) based on KEA with an applied external load (which is a fraction of the critical buckling load so that the
buckling load factor will be more than 1) to generate forces in the structural elements, which in turn provides input
for the computation of KGAKE. Then, a buckling analysis is performed effectively finding the instances when
factored by λ, KGAKE balances KEA. Hence, since both the elastic stiffness matrix KEA and the geometric stiffness
matrix KGAKE are based upon the initial undeflected (by the collapsing load factor) configuration A, the method is
only valid if deflections are small enough for there to be negligible change in KEA and the geometric stiffness
matrix varies linearly with the load factor such that it can be expressed as λKGAKE until elastic instability has
been attained. The inherent assumption that the geometric stiffness matrix varies linearly with the load factor
λKGAKE enables any external load to be applied in the first pass since KGAKE will be scaled accordingly. Hence, any
initial load (less than the collapse load so that λ < 1) may be applied and the corresponding load factor λ will be
obtained such that the collapse load λPinitial or rather expressed as λPnA will be the same. The value λ are the scale
178
Structural Engineering Analysis and Design
Second Edition
factor by which the applied load PnA is multiplied to produce the critical buckling load Pcr. The magnitude of the
applied load PnA is arbitrary for arriving at the correct Pcr. As an example, if PnA is increased by a factor of 10, then
the calculated value of λ is reduced by a factor of 10; in other words, their resulting product Pcr remains the same.
This is because the differential stiffness of the finite elements are linear functions of the axial load Fx and the
geometry l of the element (as shown below for the CBAR element) and for linear analyses, the axial load increases
linearly with the applied loads PnA.
Clearly, assuming that deflections are indeed small, the method is only valid for the first buckling mode as
subsequent deflections will certainly be too large. To illustrate the assumption, this linearized buckling analysis
basically traces the following equilibrium path.
λ
ML, GL Static Analysis
ML, GL (KGA From KEA)
Buckling Analysis λECR
[ ] {P }
∆{U} = λ K AE
−1 A
n
That is to say, the structure follows the equilibrium path governed by the constant stiffness KEA until suddenly it
loses all its stiffness (in a direction dissimilar to that of the critical load) and buckles when λKGAKE matches KEA.
This is valid if KEA is assumed not to vary with the {U} and if the component forces {f} are approximately linear in
terms of the load factor λ such that the geometric stiffness matrix varies linearly with the load factor allowing it to
be expressed as λKGAKE until elastic instability has been attained, both of which are quite valid as long as the
tangential increment of displacement due to λ is small. This can be estimated from
[ ] {P }
∆{U} = λ K AE
−1 A
n
Hence, the forces at linear buckling are the load
factor times the linear static displacement.
where ∆{U} is the tangential increment of displacement due to λ. From the expression above, clearly the same
elastic buckling displacement (and force) response will be obtained irrespective of the initial loads applied as λ will
be obtained to scale the responses accordingly. Hence, in this approach, the response at buckling is insensitive to
the initial loads applied. Conversely, calculating the displacement response from a P-∆ static analysis, we will find
that the initial loads applied do affect the response at buckling. The P-∆ response is obtained from
179
Structural Engineering Analysis and Design
Second Edition
where KGA which depends on the initial loads is obtained from a first pass
Of course, P-∆ response calculations become inaccurate when the initial load {Pm} applied is very high such that
had a linearized buckling analysis been performed, the elastic buckling load factor λ would be less than 4.
Both the buckling load factor scaled response and the P-∆ response is not ideal. In reality of course, the gradual
application of the load would cause a gradual reduction of stiffness (as the differential stiffness gradually becomes
more negative) until buckling occurs as depicted on the smoothed curve above. This path can be traced by
performing a nonlinear static and buckling analysis (MSC.NASTRAN SOL 106) and is the recommended approach
to determining the response (displacement and element forces) at buckling.
The above (KGA From KEA) may prove inadequate in many instances such as when there is prestress in the
elements even before the application of the collapsing load factor. In this case, the (KGA From KTA) elastic
buckling analysis is performed where at collapse, the tangent stiffness matrix at the initial undeflected (by the
collapsing load factor, but includes the stiffness contribution of prestressed members) stage A configuration is
balanced by the geometric (differential) stiffness due to the load factor.
[[K T ]+
A
λ[ K Gn
AKT
]
] {φ} = {0}
AKT
KGn represents the geometric stiffness matrix which was calculated based on the small displacements obtained
by solving the system with a linear static analysis (with the collapsing load) with stiffness KTA. To obtain KTA, to be
theoretically exact, a GNL SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-
DYNA) with prestress (as temperature loads say) and gravity must be undertaken. Alternatively, an approximation
to KTA can be obtained by repetitive P-∆ static analyses with the prestress (as temperature loads say) and gravity
applied. The procedure to obtain this approximate KTA will be presented. Note that the approximate KTA will be the
summation of the elastic stiffness KE at the undeflected (by the prestress and gravity) state but KG at the deflected
(by the prestress and gravity) state. Hence if KE changes considerably during the application of the prestress and
gravity, a full SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-DYNA), which
converges to the KE and KG at the deflected (by the prestress and gravity) state should be employed. Hence for the
modelling of a suspension bridge where there is a great change in geometry, it may be prudent to employ SOL 106
(or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-DYNA), but for a high tension low sag
cable on say a tower with prestressed cables, the repetitive P-∆ static analysis may be adequate. The repetitive P-∆
analysis basically involves a number of iterations of linear static analyses to obtain an approximate KTA. Note again
that A refers to the initial undeflected (by the collapsing load) state, but deflected by the prestress and gravity. To
perform the repetitive P-∆ analysis, a static analysis is performed based on KEA with temperature loads and gravity
180
Structural Engineering Analysis and Design
Second Edition
to generate forces in the structural elements, which in turn provides input for the computation of KGiAKTm where m
is the iterations. Repetitive static analysis is performed with the prestress and gravity updating the stiffness matrix
KEA + KGiAKTm-1 + KGiAKTm until convergence of displacements is obtained. The tangent stiffness at this stage is the
approximate converged tangent stiffness matrix KTA = KEA + KGiAKT. The converged displacements represent the
approximate P-∆ (KGA From Approximate KTA) static response to the initial prestress loads and gravity. The
converged geometric stiffness at this stage would be that based upon the approximate tangent stiffness matrix KTA,
i.e. KGiAKT. Irrespective of the method of obtaining KTA, whether by SOL 106 (or implicit dynamic relaxation SOL
129 or explicit dynamic relaxation LS-DYNA) or repetitive P-∆, another static analysis is performed on KTA, this
time with the collapsing applied external loads (which is a fraction of the critical buckling load so that the buckling
load factor will be more than 1). This generates KGnAKT. Then, a buckling analysis is performed, effectively finding
the instances when factored by λ, KGnAKT balances KTA. Hence, since both the tangent stiffness matrix KTA and the
geometric stiffness matrix KGnAKT are based upon the initial undeflected (by the collapsing load factor)
configuration A, the method is only valid if deflections are small enough for there to be negligible change in KTA
and the geometric stiffness matrix varies linearly with the load factor such that it can be expressed as
λKGnAKT until elastic instability has been attained. The inherent assumption that the geometric stiffness matrix
varies linearly with the load factor λKGnAKT enables any external load to be applied in the first pass since KGnAKT
will be scaled accordingly. However, in any FE model, even an extremely linear one, the geometric stiffness matrix
does not actually vary perfectly linear with the load factor, hence choosing a different external load value as a first
pass will result in minor discrepancies in the load factor. Clearly, assuming that deflections are indeed small, the
method is only valid for the first buckling mode as subsequent deflections will certainly be too large. To illustrate
the assumption, this linearized buckling analysis basically traces the following equilibrium path.
λ
ML, GL Static Analysis
ML, GL (KGA From KTA)
Buckling Analysis λECR
That is to say, the structure follows the equilibrium path governed by the constant stiffness KTA until suddenly it
loses all its stiffness (in a direction dissimilar to that of the critical load) and buckles when λKGnAKT matches KTA.
This is valid if KTA is assumed not to vary with the {U} and if the component forces {f}n are approximately linear
in terms of the load factor λ such that the geometric stiffness matrix varies linearly with the load factor allowing it
to be expressed as λKGnAKT until elastic instability has been attained, both of which are quite valid as long as the
tangential increment of displacement due to λ is small. This can be estimated from
[ ] {P }
∆{U} = λ K TA
−1 A
n
where ∆{U} is the tangential increment of displacement due to λ. In reality of course, the gradual application of the
load would cause a gradual reduction of stiffness (as the differential stiffness gradually becomes more negative)
until buckling occurs as depicted on the smoothed curve above. This path can be traced by performing a nonlinear
static and buckling analysis (MSC.NASTRAN SOL 106).
Buckling analyses based on KTA is used in instances where there is prestressing and gravity before the application
of the critical buckling loads, for instance in the buckling of suspension bridge piers or cable prestressed towers.
A static solution based upon KTA (incorporating the prestress and gravity) which is obtained exactly by SOL 106
181
Structural Engineering Analysis and Design
Second Edition
(or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation) or approximately by repetitive P-∆
analysis (the latter may be valid for low sag high tension cable problems where KE does not change much due to
the prestress and gravity) must be first obtained. Then, the buckling load is applied to generate the geometric
stiffness matrix (for that buckling load) and an eigenvalue analysis is performed.
The nonlinear (in that it is dependent upon {U}) tangent stiffness matrix is restated as follows.
[K T ] = [K E ] + [K G ]
where [K E ] = [T ]T [[k E ]][T ]
for which [k E ] = ∫ [B]T [D][B]dΩ
Ω
∂2 d
and [K G ] = [T]T [[k G ]][T] + {f }
∂{U}∂ U
∂2 ε ∂ 2 Wi ∂ 2 Wn
for which [k G ] = ∫ {{σ}n + {σ}i − [ D ]{ε}i }dΩ − − λ
∂ {d}∂ d
Ω ∂ {d}∂ d ∂ {d}∂ d
and {σ}n = [ D ][ B ]{d}
and {f } = [∫ [B] [D ][ B]dΩ ]{d} + {∫ [B] {σ} dΩ}− {∫ [B] [D ]{ε} dΩ}− {∫ [ N ] {b}dΩ}
Ω
T
Ω
T
i
Ω
T
i
Ω
Τ
As stated, we have omitted the second order variation of work due to the external nodal loadings {P} from the
above tangent stiffness matrix expression, i.e.
∂2W ∂{P} ∂W
− = − as {P} =
∂{U}∂ U ∂ U ∂{U}
This is because, as mentioned, the externally applied loads at the nodes are in commercial codes always work-
conjugate with the nodal DOFs {U}, failing which the above term must be incorporated. The second order variation
of work due to loads applied on the finite elements is obviously still taken into account.
[KT] is inherently nonlinearly dependent upon λ and {U}. The buckling problem can be expressed as a linear
eigenvalue problem if [KT] is assumed not to vary with the {U} and if the component forces {f} are
approximately linear in terms of the load factor λ. The objective is to linearize the eigenvalue problem around
the initial equilibrium state A. The first order change in [KT] for state A
∂2 d A
∂2W A
A
[ ] ∂2 d
{f } − ∆ ∂{U∂ }W ∂ W
2 2
∆ KG = ∆{f } − ∆λ n + ∆ A
− λ ∆
i A
n
∂{U}∂ U ∂{U}∂ U ∂{U}∂ U ∂ U ∂{d}∂ d
Although the above can be used directly for formulating a linear eigenvalue problem, the influence of a change of
geometry on [KT] is normally ignored, and hence ∆[KE] and the last 3 components of ∆[KG] are ignored. The
linearized buckling (eigenvalue) problem is thus expressed as
[K T ]{φ} = {0}
([K ] + λ[K ]){φ} = {0}
A
T
A
Gn
182
Structural Engineering Analysis and Design
Second Edition
The nontrivial solution is obtained by solving for the eigenvalue λ in the expression
DET ([K T ]) = 0, the corresponding {φ} being the eigenvector or the buckling mode shape
([ ] [ ])
DET K TA + λ K Gn
A
=0
[ ] = [K ] + [K ]
K TA A
E
A
G
∂2 ε A A ∂2 d A
=T [ ] [ ][T ] + [T ]
A T
k AE A A T
∂{d}∂ d
{{ }σ An + {σ}iA − [ D]{ε}iA } −
∂ 2 Wi
∂{d}∂ d
[ ]
TA +
∂{U}∂ U
{f }
i
A
∂2 ε A
2 A
∂2 d A
[K ]A
Gn =
∂{d}∂ d
{σ}An −
∂ Wn
∂{d}∂ d
+
∂{U}∂ U
{f }
A
n
where
[ ] {P }
∆{U} = λ K TA
−1 A
n
The physical meaning of the [KTA] matrix is that it represents the tangent stiffness matrix at the initial undeflected
configuration with the effect of the prestress and the initial second order work variation included. In the linearized
buckling analysis, a load factor is found which causes the tangent stiffness to become zero by a term which
multiplies this load factor to the nominal part of the geometric (or differential) stiffness based on the initial
undeflected configuration. The nominal part refers to the additional element force {fn} and additional work Wn
terms, not the prestress {fi}. Hence, it is assumed that in the application of the load factor that both the nominal
internal forces {fnA} and the nominal second order work term vary linearly with the load factor. These factored
nominal internal force and factored nominal second order work terms are based on values at the initial state A. Also
it is assumed that the transformation matrix [T] is constant at [TA]. Thus it can be concluded that the linearized
buckling analysis is only valid when the deflections {U} are small i.e. to predict buckling along the trivial
fundamental path or along a path close to the trivial path.
A buckling point in the response of a discrete structural system is either a bifurcation point or a limit point.
Accordingly, it corresponds to an equilibrium state for which the tangent stiffness matrix [KT] becomes singular,
indicating the presence of an infinitesimally adjacent equilibrium state at the same level of loading. Linearized
buckling analysis is based on the initial undeflected geometry, i.e. (A) = initial state to determine [KTA]. [KTA] is
thus the tangential value of [KT] at the initial undeflected shape (A). There is no point in choosing another state (A)
to find [KTA], as we then might as well perform nonlinear buckling analysis.
Linear buckling analysis tends to provide good approximations to the critical load when the critical point is a
bifurcation point. But linear buckling analysis can overestimate the critical load when the critical point is a limit
point and if the buckling occurs at significant incremental deflection ∆{U} from state (A) as the change in [KT] due
to the change in {U} is ignored. This value of ∆{U} is estimated using the equation stated above in order to make a
judgment. Higher critical modes in linear buckling analysis are inaccurate if the first mode is a limit point due to
subsequent larger deflections. Hence, linear buckling analysis is most suitable for detecting bifurcation points along
trivial (or almost trivial) fundamental paths, in which case state (A) can be the initial state, and for detecting critical
points which are close to the known equilibrium state (A).
183
Structural Engineering Analysis and Design
Second Edition
The number of degrees of freedom n of the linear eigenvalue problem can be reduced to m as an approximation.
The linear eigenvalue problem of size n is expressed as
([K ] + ∆λ[K ]){Φ} = {0}
A
T
A
Gn
In the special case that the trial mode is to be approximated straight away such that the trial mode matrix [Φ t ]
reduces to a single column vector {Φ t }, then
{Φ} = {Φ t }{φ}
for which the approximate eigenvalue ∆λ is obtained from
∆λ = −
k TA
= −
[ ]
{Φ t }T KTA {Φ t }
A
k Gn {Φ t }T K Gn
A
[ ]
{Φ t }
and {Φ t } being the corresponding buckling modes shape or eigenvector
A good approximate mode shape for uniform shear frames based on Rayleigh's is {Φ t } = K TA [ ] {P }
−1
n
where {Pn } is the vertical loads applied horizontally.
Naturally, the solution would be exact if {Φ t } is the exact mode shape. This is ensured if
[K ]{Φ } = λ[K ]{Φ }
A
T t
A
Gn t i.e. [K ]{Φ } α [K ]{Φ }
A
T t
A
Gn t
The trial mode matrix [Φt] for the problem reduction in linear buckling analysis is a multi modal matrix. The trial
mode method assumes just one mode shape {Φt} whilst the problem reduction method assumes a linear
combination of mode shapes, [Φt] = [{Φt1}+{Φt2}+{Φt3}+…+{Φtm}]. Hence the problem reduction method is more
accurate evidently. Hence, the problem reduction method will yield a lower estimate of ∆λ than the trial mode
shape method. The nonlinear eigenvalue problem will of course yield the lowest exact value of ∆λ.
184
Structural Engineering Analysis and Design
Second Edition
The concept of GL, ML buckling analysis will be illustrated for a simple system.
Let us assume that deflections are small (hence geometrically linear analysis) hence the rotation of the column
segments can be described and any change in height can be neglected. Lateral equilibrium at the central pin shows
that
Rearranging so that we have the form Stiffness x Displacement = Applied Force, we get
We can see that the effect of the force N acting on this system can be seen as being like an additional stiffness term
2N/L, i.e. the geometric (aka differential) stiffness. The effect of the axial force is equivalent to an additional
stiffness of value 2N / L acting laterally at the pin location. If the force is tensile (i.e. positive) then the geometric
stiffness will act to increase the overall stiffness of the system. On the other hand, if it is compressive then the term
will be negative and the system stiffness will decrease. Consider the condition at which the axial compression in
our simple pin jointed system is so great that the lateral stiffness of the system is zero.
i.e.
At this critical load, buckling occurs. And the critical linear elastic buckling load factor λcr of the system is
It cannot be emphasized enough that, the buckling load factor is independent
of the initial lateral imperfections, the lateral load and material strength.
This linear buckling critical load factor is independent of the applied lateral force F and the initial lateral
imperfection. Of course, the amplified displacement response is dependent upon F and the initial imperfection.
The buckling factor is also independent of the material strength as it depends on the stiffness instead.
185
Structural Engineering Analysis and Design
Second Edition
SOL 105
$ CASE CONTROL SECTION
DISPLACEMENT(<PRINT,PUNCH,PLOT>) = ALL/<Grid Set ID> $ Plots the buckled shapes
SPCFORCES(<PRINT,PUNCH,PLOT>) = ALL/<Grid Set ID>
OLOAD(<PRINT,PUNCH,PLOT>) = ALL/<Grid Set ID>
ELSTRESS(<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
ELFORCE(<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
STRAIN(<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
ESE(<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
$
SUBCASE 1
LABEL = Static Load Case With Buckling Load
LOAD = < ID of FORCEi, MOMENTi, PLOADi, GRAV, LOAD, SPCD Cards in Bulk Data >
TEMP(LOAD) = < ID of TEMP, TEMPRB, TEMPP1 Cards in Bulk Data >
DEFORM = < ID of DEFORM Cards in Bulk Data >
SPC = < ID of SPC Cards in Bulk Data >
$
SUBCASE 2
LABEL = Buckling Case
STATSUB(BUCKLING) = < Static Subcase Number >
SPC = < ID of SPC Cards in Bulk Data >
METHOD = < ID in EIGRL or EIGB >
For medium to large models, the Lanczos method is the recommended eigenvalue extraction method. Furthermore,
the Lanczos method takes full advantage of sparse matrix methods that can substantially increase computational
speed and reduce disk space usage. For overall robustness, the Lanczos method is the recommended method.
$ BULK DATA
Unlike dynamic modal analysis, linear buckling analysis produces eigenvalues that can be either positive or
negative. Note that these negative roots are legitimate both from a physical and a mathematical standpoint. In
general, this situation only occurs if the applied buckling load is in the opposite direction of the buckling load, that
is to say a buckled shape with the same magnitude of eigenvalue can be produced if the applied loading was
reversed in direction. Hence, if on the EIGRL card, only the upper eigenvalue range is specified, NASTRAN will
attempt to extract all eigenvectors below that value, including all the negative eigenvalues. Thus if we are only
interested in the fundamental mode, it is necessary to also specify the number of roots.
Buckling load factor, λ
The enhanced inverse power method (SINV) can be a good choice if the model is too large to fit into memory, only
a few modes are needed, and you have a reasonable idea of your eigenvalue range of interest. It is useful for models
186
Structural Engineering Analysis and Design
Second Edition
in which only the lowest few modes are desired. This method is also useful as a backup method to verify the
accuracy of other methods.
$ BULK DATA
Lower Upper
EIGB ID SINV
Eigenvalue Frequency
At least two subcases are necessary, the first being the static subcase that is used in the second subcase to generate
the differential stiffness matrix from the internal element loads and then solve the eigenvalue problem. The
differential stiffness is supported by CONROD, CROD, CTUBE, CBAR (for both Euler and lateral torsional
buckling), CBEAM, CBEND, CQUAD4, CQUAD8, CTRIA3, CTRIA6, CSHEAR, CHEXA, CPENTA and
CTETRA. The contribution to the differential stiffness from the follower force effect is not included. Offsets
should not be used in beam, plate, or shell elements for buckling analysis. For 3-D buckling problems, the use of
PARAM, K6ROT is recommended for CQUAD4 and CTRIA3 elements. A value of 100 is recommended. Note
that each subcase may have a different boundary condition.
Although the element stresses, forces and strains for each buckled mode may be output, it is meaningless as they
are based upon the eigenvectors that have been scaled arbitrarily. These values DO NOT INDICATE
RESPONSE. To obtain the response (displacement and element forces) at buckling, the response from a linear
static analysis of the initial applied loads, and scaled by the critical load factor can be used.
[ ] {P }
∆{U} = λ K AE
−1 A
n
Of course though, the element strain energy density distribution (ESE request), although have likewise meaningless
magnitudes (again due to arbitrary scaling of eigenvectors), is an important design tool which shows elements
which work the most in causing that buckled mode shape. Stiffening elements with a high ESE would increase the
buckling load factor.
The number of buckled modes obtained will be limited to the number of DOFs in the finite element model. The
mesh should be fine enough to represent the buckled shape. CBAR elements have cubic polynomial transverse
interpolation functions (shape functions). But unlike in a static analysis where the cubic deflection over a beam
element can be mapped, in a buckling analysis the deflected shape is provided only by the deflections at the nodes.
Hence the beam element must be refined. Also note that the buckled mode shape of a beam member is not cubic
polynomial anyway. One dimensional elements such as CBAR elements (in a fine mesh as they may be) do not
model the local buckling of members. The local buckling of a tube or pipe (i.e. the crippling of the thin wall) in
compression must be modeled with a fine mesh of shells. A coarse mesh may predict an unconservative larger
fundamental buckling load factor as the local mode cannot be captured. It is thus important to anticipate the
buckling mode, or alternatively, continuous refinement be made. Buckling is a state of loss of stiffness, an unstable
state, i.e. failure. If the local crippling buckling mode is more critical than the overall beam buckling mode, it is
imperative that shell elements are used as line elements will give an unconservative result. Buckling generally
requires a fine mesh. If the buckled shape is jagged, it is a clear indication that the mesh is not fine enough. There
must be several elements per wavelength, usually 4 (or 8, 16, 32 etc…other numbers even 5 tend to be inferior)
elements per half sine wavelength.
Local buckling can also be checked using hand calculations based upon the physical boundary conditions and
thickness on a relevant panel to estimate the buckling stress and comparison with the stress from a linear static SOL
101 analysis to evaluate if the component would buckle under that load.
Linear buckling analysis assumed the deflections to be small and the elements stresses elastic. For structures that
exhibit nonlinear material or large deflection deformations, the linear buckling load obtained from Solution 105
may be different than the actual buckling load. For structures with significant nonlinearities, it is recommended that
a nonlinear buckling analysis using SOL 106 be performed.
187
Structural Engineering Analysis and Design
Second Edition
The preliminary scheme stage buckling analysis of the Centre Pompidou, Metz timber roof is presented.
The applied load consists of timber planks (6097kN), intermediate packs (1039kN) and roof covering (2411kN).
The linearized buckling problem is
[[K E ]+
A
]
λ[ K GAKE ] {φ} = {0}
The instantaneous and geometric stiffnesses are based on the unfactored 5th percentile stiffness values (E = 12GPa;
G = 0.4GPa) for the timber material, hence conservative as far as buckling analysis is concerned. The geometric
stiffness is a function of the internal forces. For example, for a straight column subject to axial compression F, the
geometric stiffness is simply F/L. The critical load case which causes the greatest compression and hence the
greatest geometric stiffness is sought.
The lowest buckled mode shape (which could be replicated in reality) due to self-weight is presented. The load
factor is 16.9. Note that there was another mode with a load factor of –14.75, this indicating an upward gravity,
which is implausible, hence ignored.
Hence the first global buckling load factor to unfactored self-weight = 16.9. From plot, buckled mode shape shows
primarily vertical displacement, as the normalization is 1000 mm and the vertical displacement reach up to 982.7
mm.
The static displacement at the onset of buckling can be estimated by the following projection
[ ] {P }
∆{U} = λ K AE
−1 A
n
This can be used to ascertain whether geometrically nonlinear effects or material nonlinearity for that matter,
affects the estimation of the buckling load and the static displacement at the onset of buckling. The following is the
static displacement from the unfactored self-weight load case.
188
Structural Engineering Analysis and Design
Second Edition
The predicted max static displacement from the unfactored self-weight case at onset of buckling is
16.9 x 143mm = 2414mm
Note that this is the max displacement, not the average. A geometrically nonlinear buckling analysis is required to
predict a more accurate buckling load.
The factor of safety on buckling to the ULS load case is of interest. The first realistic mode shape is presented. The
load factor is 6.33.
As a check, from a linear projection of the self-weight case, since the distribution of load is similar, the load factor
from this ULS case should be 9547 / 22530 * 16.9 = 7.15, not too dissimilar to the above prediction.
The following is the static displacement from this ULS load case.
189
Structural Engineering Analysis and Design
Second Edition
Hence, at the onset of buckling from the ULS case, from the above projection, the max static displacement is
6.33 x 382mm = 2418mm
This is a 2.4m deflection in a span of 40m, i.e. span/depth of 16.5. The effect of this on the accuracy of the linear
buckling load factor estimate needs to be quantified using a geometrically nonlinear buckling analysis solver.
The North Wind load case refers to an asymmetric wind load case such that the windward direction experiences a
downward pressure and the leeward direction experiences an upward pressure. The first realistic mode shape is
presented. The load factor is 5.15.
Conclusion
190
Structural Engineering Analysis and Design
Second Edition
The critical self-weight linearized buckling load factor is 16.9. Since the value is greater than 10.0, the effect of P-∆
in increasing the effects on the structure and reducing the load factor due to self-weight is negligible. The critical
ULS linearized buckling load factor is 5.15. Since the value is less than 10.0, the effect of P-∆ in increasing the
effects on the structure and reducing the load factor due to the ULS load case is not negligible. However the risk of
catastrophic buckling instability is low.
Assumptions
2. Geometrically linear; Elastic stiffness KEA constant until onset of buckling and based on initial
undeformed geometry; Valid if deflections small (for negligible change in KEA) until onset on buckling and
no prestress (whether compressive which is unfavourable or tensile which is favourable) in structure.
Check magnitude of ∆{U} to be small and ensure no prestress in system.
3. Geometrically linear; Geometric stiffness KGKEA based on KEA; Valid if deflections small (for negligible
change in KEA and KGKEA) until onset on buckling. Check magnitude of ∆{U} to be small.
191
Structural Engineering Analysis and Design
Second Edition
The above subcase definitions will solve the following eigenvalue problem.
[[K E ]+
A
]
λ[ K GAKE ] {φ} = {0}
AKE
KG represents the geometric stiffness matrix which was calculated based on displacements obtained by solving
the system with stiffness KEA. This may prove inadequate in many instances such as when there is prestress in the
elements. The linearized buckling problem as described in the theory section is sought as
[[K T ]+
A
λ[ K Gn
AKT
]
] {φ} = {0}
AKT
KG represents the geometric stiffness matrix which was calculated based on displacements obtained by solving
the system with stiffness KTA.
To obtain KTA, to be theoretically exact, a GNL SOL 106 (or implicit dynamic relaxation SOL 129 or explicit
dynamic relaxation LS-DYNA) with prestress (as temperature loads say) and gravity must be undertaken.
Alternatively, an approximation to KTA can be obtained by repetitive P-∆ static analyses with the prestress (as
temperature loads say) and gravity applied. The procedure to obtain this approximate KTA will be presented. Note
that the approximate KTA will be the summation of the elastic stiffness KE at the undeflected (by the prestress and
gravity) state but KG at the deflected (by the prestress and gravity) state. Hence if KE changes considerably during
the application of the prestress, a full SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic
relaxation LS-DYNA), which converges to the KE and KG at the deflected (by the prestress and gravity) state
should be employed. Hence for the modelling of a suspension bridge where there is a great change in geometry
(known in the bridge industry as form-finding, so-called because it is necessary to find the form or shape of the
catenary suspension cables), it may be prudent to employ SOL 106 (or implicit dynamic relaxation SOL 129 or
explicit dynamic relaxation LS-DYNA), but for a high tension low sag cable on say a tower with prestressed
cables, the repetitive P-∆ static analysis may be adequate. The repetitive P-∆ analysis basically involves a number
of iterations of linear static analyses to obtain an approximate KTA. Note again that A refers to the initial
undeflected (by the collapsing load) state, but deflected by the prestress and gravity. To perform the repetitive P-∆
analysis, a static analysis is performed based on KEA with temperature loads and gravity to generate forces in the
structural elements, which in turn provides input for the computation of KGiAKTm where m is the iterations.
Repetitive static analysis is performed with the prestress and gravity updating the stiffness matrix KEA + KGiAKTm-1 +
KGiAKTm until convergence of displacements is obtained. The tangent stiffness at this stage is the approximate
converged tangent stiffness matrix KTA = KEA + KGiAKT. The converged displacements represent the approximate P-
∆ (KGA From Approximate KTA) static response to the initial prestress loads. The converged geometric stiffness at
this stage would be that based upon the approximate tangent stiffness matrix KTA, i.e. KGiAKT.
Phase 1
Perform static analysis (with prestress and gravity) based on [KEA] but include the segyroa.v2001 alter prior to the
Case Control Section in order to compute the geometric stiffness matrix [KGA]1 (and output into a .pch file) based
on the generated element loads from the [KEA] static analysis.
Phase 2
Perform static analysis (with prestress and gravity) based on [KEA] + [KGA]1 by including the k2gg = ktjj statement
in the Case Control Section, the outputted .pch file which contains the ktjj matrix in the Bulk Data and the
segyroa.v2001 alter prior to the Case Control Section to compute the [KGA]2 (and output into the .pch file
overwriting previous data) based on the generated element loads from the [KEA] + [KGA]1 static analysis.
Phase 3
Repeatedly perform the Phase 2 static analysis (with prestress and gravity) based on [KEA] + [KGA]i for i = 2 to n
where n represents the number of iterations required for the change in deflections between analyses to become
192
Structural Engineering Analysis and Design
Second Edition
negligible. This would signify that the change in the [KGA] matrix become negligible and the correct [KGA] is
attained. The deflections and the other responses at this stage represent the P-∆ (KGA From Approximate KTA) static
response to the prestress and gravity. The stiffness of the structure is KTA.
Phase 4
Linearized buckling analysis is now performed with the stiffness of the structure at KTA and with a general buckling
load. A SOL 105 is undertaken with the Case Control Section described as follows. It includes one more static
analysis with the collapsing buckling load (a fraction of the critical load so that the load factor is more than 1) in
order to generate the KGnAKT matrix. Then a buckling analysis is performed.
SUBCASE 1
LABEL = P-∆ (KGA From Approximate KTA) Static Load Case With Buckling Load
k2gg = ktjj
LOAD = < ID of FORCEi, MOMENTi, PLOADi, GRAV, LOAD, SPCD Cards in Bulk Data >
TEMP(LOAD) = < ID of TEMP, TEMPRB, TEMPP1 Cards in Bulk Data >
DEFORM = < ID of DEFORM Cards in Bulk Data >
SPC = < ID of SPC Cards in Bulk Data >
SUBCASE 2
LABEL = Buckling Case
k2gg = ktjj
SPC = < ID of SPC Cards in Bulk Data >
METHOD = < ID in EIGRL >
$ BULK DATA
The responses at this stage represent the (KGA From Approximate KTA) linear buckling response.
We have thus introduced so far 4 methods of getting the KTA matrix, namely full SOL 106, an implicit dynamic
relaxation SOL 129, an explicit dynamic relaxation LS-DYNA or a repetitive P-∆ static analysis, all with the
prestress and other loads. Once the KTA matrix is obtained, all that is required is to perform a linear static analysis
with the buckling loads in order to generate a KGA (clearly from KTA) matrix. Now the linearized buckling problem
can be solved as the linear eigenvalue problem [KTA + λKGA]{φ}={0}. Another approximate procedure can also be
employed in order to generate the KTA matrix.
193
Structural Engineering Analysis and Design
Second Edition
The method is valid when only the prestress is judged to affect the geometric stiffness such as in the
compressive preload of building columns due to gravitational loads and the prestressing of extremely taut cables
that sag very little under gravity but not in systems such as suspension bridges. Where lateral loads are large
enough to affect the geometry of the system with prestress, then a repetitive P-∆, SOL 106, implicit dynamic
relaxation SOL 129 or explicit dynamic relaxation must be employed. But in single P-∆ analysis, because cables do
not have much elastic bending stiffness, the initial static preload subcase should only include the prestress and not
gravity as including gravity is the same as solving two linear static problems of stiffness KEA with preload and
gravity as the applied loads respectively. Clearly, in the gravity case, it is nonsensical as the cables do in reality
have differential stiffness (from the prestress) to resist the gravitational force. Prestress in one direction (i.e. along
the axis of cable) will cause a differential stiffness in the orthogonal direction. Gravity acts in the orthogonal
direction and hence cannot be accounted for in the calculation of the prestress in this single P-∆ analysis. To
quantitatively decide if gravity need not be considered in contributing to the differential stiffness of the cables, a
static P-∆ analysis should be carried out, the first subcase being a SOL 101 with only the prestress as applied loads
and the second subcase a P-∆ SOL 101 (i.e. utilizing the induced prestress from the first subcase to form a
geometric stiffness matrix) with both the gravity and prestress included as applied loads. If the difference in the
cable element forces between subcases 1 and 2 is negligible, then gravity has little influence in affecting the
geometric stiffness. If there is a major difference in the cable element force, then clearly, gravity will affect the
geometric stiffness and as such, a repetitive P-∆ , SOL 106, implicit dynamic relaxation or explicit dynamic
relaxation must be used to converge to the true KT. Likewise, in the single P-∆ analysis of multi-storey buildings,
gravity (and only gravity) acts in axis of columns to generate prestress, and the differential stiffness is computed for
the orthogonal direction reducing resistance to lateral wind forces, applied in the second subcase with gravity too.
The STATSUB(PRELOAD) computes the differential stiffness due to the prestress and also the follower force. The
follower force is calculated and incorporated by the use of PARAM, FOLLOWK, YES. We know how the prestress
affects the differential stiffness, namely a tensile prestress causing an increase in stiffness. The effect of the follower
force on the stiffness is different. For example, for a cylinder under external pressure critical buckling load may be over-
estimated (even though the mode shapes are similar) in a SOL 105 and the natural frequencies in vibration may be under-
estimated (even though the mode shapes are similar) in a SOL 103 in the absence of follower stiffness. To the contrary,
these observations are reversed in case of centrifugal loads. Centrifugal forces as a constant (static) load are applied by a
Bulk Data RFORCE to any elements that have masses. The follower stiffness due to centrifugal load has the effect of
lowering stiffness (although the centrifugal load tensioning effect increases stiffness), consequently lowering natural
frequencies (even though the mode shapes are similar) in a SOL 103 and lowering the buckling loads (even though the
mode shapes are similar) in a SOL 105. This effect increases as the RPM increases, and it becomes significant when the
RPM is over 1000.
Another important observation is that the preload cannot buckle the structure. The basic eigenvalue problem was
expressed as
where λ is an eigenvalue which is a multiplier to the applied load to attain a critical buckling load. If there exists
constant preloads other than the buckling load in question, the above equation should include additional differential
stiffness, i.e.
in which differential stiffness is distinguished for constant preload and variable buckling load. Notice that no
eigenvalue solutions are meaningful if the preload makes the structure buckle, i.e.
194
Structural Engineering Analysis and Design
Second Edition
3.2.5.1.1 ML, GL Euler Buckling of Column Elements (Primary Stress is Only Axial)
The Euler equation for the critical elastic buckling load for an elastic, isotropic and straight column is
π 2 EI The Euler load estimate is perfectly analogous
PE = 2
LE to the SOL 105 buckling load estimate.
where LE is the effective length (distance between the points of contraflexure) depending on the boundary
conditions.
Boundary Conditions Effective Length, LE
Non-Sway Sway
Pinned-Pinned 1.0L (Truss)
Pinned-Rigid 0.699L 2.0L
Rigid-Rigid 0.5L 1.0L
The Euler equation is of course only valid for columns that are sufficiently long enough to experience linear elastic
instability before yielding. Hence it is better to work in terms of axial stresses as the yield stress is clearly defined.
The critical Euler stress is
π2E I
σE = where r =
(L E / r ) 2
A
It cannot be emphasized enough that, the
and is only valid if theoretical buckling load factor is independent
of the initial lateral imperfections, the lateral
LE π2E load and material strength.
>
r σy
This is all illustrated in the following diagram.
σ cr The curves are fully described by the Rankine
σ cr = σ y σy
σy buckling formula σ RANKINE =
π2E σy L 2
σ cr = 1+ 2 e
(L E / r )2 π E r
LE
Slendernes s
r
LE π2E
=
r σy
The buckling stress capacity depends on the stiffness of the element and the boundary conditions, both of which is
accounted for in the slenderness parameter LE/r. For low slenderness, the beam fails by yielding, and for high
slenderness, it fails by buckling. This slenderness parameter should be checked for both the major axis and the
minor axis as the boundary conditions may differ. The capacity corresponding to the greater slenderness is clearly
critical. The critical load capacity is
PE = σ E A
The Euler buckling load presented hitherto is actually the flexural Euler buckling load
π 2 EA
PE = σ E A =
(L E / r )2
195
Structural Engineering Analysis and Design
Second Edition
There is also the much smaller possibility of shear buckling. When buckling occurs, shear forces act on the section.
These give rise to additional shear deformations that reduce the force that the member can take in compression
before it becomes unstable and buckles.
The critical load for combined flexural and shear buckling is given by
1/Ncrit = 1/Ncrit.flexural +1/Ncrit.shear
Hence, the critical force is reduced approximately by the factor 1/[1+Ncrit/GAs), where Ncrit is the critical force in
the absence of shear deformation.
On top of Euler flexural and shear buckling, columns are also subject to torsional buckling and flexural-
torsional buckling. These occur when the ratio of torsional to flexural stiffness is very much reduced.
196
Structural Engineering Analysis and Design
Second Edition
3.2.5.1.2 ML, GL Lateral-Torsional Buckling of Beam Elements (Primary Stress is Only Bending)
From classical mechanics, it can be shown that the lateral torsional buckling capacity for an elastic, isotropic beam
with a uniform bending moment diagram is
The lateral-torsional buckling of a beam is governed by both the minor axis elastic second moment of area, IMINOR
and also the torsional constant, J. Contrast with the Euler buckling which depended upon only the minor axis elastic
second moment of area, IMINOR.
The π is replaced with another value depending on the bending moment diagram; 4.24 for a triangular BMD due to
a central point load; 3.53 for a uniform uniformly distributed load BMD.
However, beams are subject to plastic failure (attainment of plastic moment capacity) at low slenderness and to
lateral torsional buckling at higher slenderness. The plastic moment capacity is
M plastic = S plastic σ y
This is all illustrated in the following diagram. It cannot be emphasized enough that, the
theoretical buckling load factor is independent
σ cr of the initial lateral imperfections, the lateral
M plastic
σ cr = = σy load and material strength.
S plastic
σy
M LTB
σ cr =
S plastic
LE
Slendernes s
r
The buckling stress capacity depends on the stiffness of the element and the boundary conditions, both of which is
accounted for in the slenderness parameter (LE/r). For low slenderness, the beam fails by yielding, and for high
slenderness, it fails by lateral torsional buckling.
Lateral torsional buckling cannot occur in beams loaded in their weaker principal plane; under the action of
increasing load they will collapse simply by plastic action and excessive in-plane deformation. Hence, the check
need only be made for bending about the major axis.
197
Structural Engineering Analysis and Design
Second Edition
Theoretical solution for the Euler buckling of a plate (panel) of thickness t, width of b and the two loaded edges
simply supported is a
kπ 2 E b
σE =
( )
12 1 − ν 2 (b / t )
2
where k is dependent upon the boundary conditions of the two non-loaded edges and also on the ratio of a/b, where
a is the length of the panel. The table below presents the minimum values of k for different boundary conditions of
the unloaded edge, with the loaded edge simply supported.
Unlike columns, plates do not collapse when the critical buckling load is achieved. Constraints due to the two-
dimensional nature of the plates allow plates to resist increasing loads beyond buckling. Hence the nonlinear post-
buckling analysis is more important for plates.
In BS 5950-Part 1:2000, cross-sections are classified to determine whether local buckling influences their
capacity, without calculating their local buckling resistance. The classification of each element of a cross-section
subject to compression (due to a bending moment or an axial force) should be based on its width-to-thickness ratio.
The section classes are as follows.
Class 1 Plastic: Sections with plastic hinge rotation capacity. The full plastic modulus S may be used in design.
Class 2 Compact: Sections with plastic moment capacity. The full plastic modulus S may be used in design.
Class 3 Semi-Compact: Sections in which the stress at the extreme compression fibre can reach the design
strength, but the plastic moment capacity cannot be developed. Elastic design methods must be adopted, i.e. the full
elastic modulus Z must be used in design.
Class 4 Slender: Sections in which it is necessary to make explicit allowance for the effects of local buckling. BS
5950-Part 1:2000 gives approximate methods using an effective area for Z calculation in cl. 3.6.
198
Structural Engineering Analysis and Design
Second Edition
Note the much tighter criteria (about 9ε, 10ε, 15ε) for local sections which are not constrained such as the outstand
of a flange, an angle or the stem of T- section. Note the fairly relaxed criteria (about 40ε, 40ε, 40ε) for partially
constrained sections such as the web of a channel. And note the very relaxed criteria (about 80ε, 100ε, 120ε) for
constrained sections such as the web of a I-, H- or Box section.
199
Structural Engineering Analysis and Design
Second Edition
200
Structural Engineering Analysis and Design
Second Edition
201
Structural Engineering Analysis and Design
Second Edition
Additional considerations to moment capacity must be made according to cl. 4.4.4.2 if web is subject to shear
buckling.
202
Structural Engineering Analysis and Design
Second Edition
203
Structural Engineering Analysis and Design
Second Edition
204
Structural Engineering Analysis and Design
Second Edition
205
Structural Engineering Analysis and Design
Second Edition
206
Structural Engineering Analysis and Design
Second Edition
207
Structural Engineering Analysis and Design
Second Edition
208
Structural Engineering Analysis and Design
Second Edition
209
Structural Engineering Analysis and Design
Second Edition
210
Structural Engineering Analysis and Design
Second Edition
211
Structural Engineering Analysis and Design
Second Edition
212
Structural Engineering Analysis and Design
Second Edition
The approximate bifurcation analysis of multi-story frames estimates an approximate elastic critical load, λECR.
BS 5950-Part 1:2000 (cl. 2.4.2.6) gives a method for calculating λcr for simple (symmetric in plan) multi-storey
buildings (excluding single-storey frames with moment-resisting joints and other multi-storey frames in which
sloping members have moment-resisting connections) based on the deflection due to the Notional Horizontal
Forces of 0.5% of the factored vertical (1.4 dead + 1.6 live) load or 1.0% of the factored vertical (1.4 dead)
load, whichever the greater applied at the same level.
1 h
λ cr =
200 δ
where h/δ is the value of the storey height divided by the deflection over that storey (deflection at top of storey
relative to deflection at bottom of storey) for any storey in the building. The smallest value considering every
storey should be used. This equation is based on a shear mode buckling and gives very good agreement with more
detailed analysis. Lower buildings tend to be governed by shear buckling whilst taller buildings are governed by the
flexural buckling mode. If the mode is flexural, the above simple equation is likely to underestimate (conservative)
λcr. The stability systems for steel buildings are often braced cores. It can be shown that if the stress in the
diagonals is greater than 57MPa, however stiff the columns and beams, λcr will be less than 10.
213
Structural Engineering Analysis and Design
Second Edition
3.3 GL, MNL Plastic Collapse Analysis by the Implicit Linear Simplex Programming
Great structural engineering failures since time immemorial can be attributed to a lack of redundancy. A plastic
collapse analysis is a brilliant method of finding out the amount of redundancy that a structure had to a load.
λ ML, GL Static Analysis
MNL, GL Plastic
Collapse Analysis λP
MNL, GNL Static and Buckling Analysis
(Tracing Equilibrium Path or Implicit or Explicit
MNL, GL Elasto-Plastic
Time Integration with Dynamic Relaxation) λF
Collapse Analysis λE
U
Plastic collapse analysis is performed if material non-linearity (yielding and subsequent formation of plastic hinges)
is thought to be more significant than elastic buckling instability in causing the failure of the structure. The
structure has not buckled (experienced elastic instability) but has α + 1 fully plastic hinges, i.e. a mechanism.
In order to use plastic collapse analysis methods, the beams and columns must be ductile, i.e. must be able to
deform sufficiently to form plastic hinges. Ductility in beams improved by having less than 4% steel reinforcement,
having shear links to confine concrete in three dimensions, having a little compression steel, too much of which
will reduce ductility. Ductility in columns improved by having shear links to confine concrete in three dimensions
to avoid the propagation of micro-cracks.
The Uniqueness Theorem states that the smallest critical plastic collapse load factor is attained if the 3 following
condition are achieved: -
I. Equilibrium condition: The bending moment distribution must be in equilibrium with the external
applied loads
II. Yield condition: The bending moment must nowhere exceed the plastic moment capacity of the section
III. Mechanism condition: There must be sufficient plastic hinges to form a mechanism.
The Upper Bound Mechanism Method of Plastic Limit Analysis to find λP uses the equilibrium conditions and
postulates many collapse mechanisms, finding the critical combination by minimizing λP. If the postulated collapse
mechanisms subsumes the critical one, then the collapse bending moment diagram will show that no part of the
structure does the moment exceed the plastic moment capacity, i.e. the yield condition would have been met.
min λ
λOPT
The proposed plastic collapse analysis is a geometrically linear method. Often, P-∆ effects (reduction in lateral
bending stiffness of axially loaded columns) need to be accounted for. We could approximately incorporate
geometric non-linearity to find an approximate λF by employing Wood’s adaptation of the Merchant-Rankine
empirical formula based on an estimate of the critical elastic load factor, λE. Note also that should the axial forces
be large in certain elements, the value of the plastic moment capacity MP would decrease in accordance with the M-
N diagram.
214
Structural Engineering Analysis and Design
Second Edition
(i) Select critical sections C or C* (i.e. whether simple frame or multi-bay multi-storey frame), identify degree
of static indeterminacy α or α* (i.e. whether simple frame or multi-bay multi-storey frame), and establish the
sign convention for positive bending moments and rotations.
- fixed supports
- joints between members
- between two members without applied moment, only one possible critical section at the joint,
i.e. at the weaker member
- between two members with applied moment, two possible critical sections
- at a 3 or more member joint, all the member ends must be included as possible critical sections
- under concentrated loads
- under uniform loads
- at changes of member detailing, i.e. discontinuity in MP and plastic hinge occurs in weaker section
Hyperstatic Isostatic
Many stress paths from any point to root Only one stress paths from any point to root
Bending moments and rotations are defined as positive if they cause tension or extension at the side with the
dotted lines. For multi-storey multi-bay buildings, the dotted line is drawn under beams and on the right of
columns by convention.
p
m C = [B b 0 ] α
λ
p
m1 B . . b0 1 B
. . . . p1 b0
. . .
= . . . .
. . .
. . .
. = + λ
. . . . . . . . . . . .
p
mC . . . B b0 α . . . B. pα b0
λ
215
Structural Engineering Analysis and Design
Second Edition
(iii) Formulate the mesh kinematic linear program for the plastic limit analysis
b T − b T0 & 1
minimise λ = M P θ&* 2C subject to T0 θ =
T
T *
B − B 0
positivity condition θ&* ≥ 0
(vii) Establish the bending moment distribution at collapse from the equilibrium equations
(viii) Establish the actual elastic and plastic deformations (i.e. rotations and displacements)
[Ba]T{θPa} = −[B]T[f]{m}
[Ba] = mesh matrix associated with active plastic hinges
{θPa} = active plastic hinges vector
[B] = complete mesh matrix
[f] = flexibility matrix
{m} = bending moment at collapse vector
{d} = [B0]T {{θE}+{θPa}}
= [B0]T {[f]{m}+{θPa}}
{d} = displacement vector
[B0] = matrix equilibrating a unit load without compatibility considerations at each and
every station where the deflection is sought
(ix) Employ the approximate bifurcation analysis for simple frames or other methods to find an approximate λECR
∑V h φ
i
2
i i
S/ V
Horne' s Estimate λ ECR = 0.9
Max φi
(x) Employ Wood’s adaptation of the Merchant-Rankine empirical formula to find an approximate λF. This
effectively accounts for the P-∆ (KGA From KEA) effect from the linear elastic buckling load, which is based
on the initial undeflected geometry.
λF 1 λ ECR
= for 4≤ ≤ 10
λP λ λP
0.9 + P
λ ECR
λF λ ECR
= 1 for ≥ 10 i.e. assume no geometricnonlinearity effect beyond ratio of 10
λP λP
Note that in conceptually well designed structures, λECR >> λP such that λF ~ λP (Merchant-Rankine). Also,
λE << λP so that the structure deforms considerably in plasticity before failure.
216
Structural Engineering Analysis and Design
Second Edition
p
m = [B B 0 ]
λ
p1
.
m1 B . . . B0 . .
. . .
= . . . . . .
p
. . . . . . . . α
λ
mC . . . B . . B0 1
.
λ
total
Incorporating the single parameter loading λ, the equilibrium law equations become
p
m = [B b 0 ] = B p + λ b 0
λ
p
m1 B . . b0 1 B
. . . . p1 b0
. . .
= . . . .
. . .
. . .
. = + λ
. . . . . . . . . . . .
p
mC . . . B b0 α . . . B. pα b0
λ
v& B T &
δ& = T ϑ
B 0
v&1 B T . . .
.
. . . . &
θ
. . . . . 1
.
v&α = . . . BT
.
δ&1 B0T . . . &
θ
. . . . . C
δ& . . . B0T
total
217
Structural Engineering Analysis and Design
Second Edition
Incorporating the single parameter loading λ, the kinematic law equations become
v& B T &
W& = T ϑ
λ =1 b 0
v&1 B T . . . &
. θ
. . . . 1
.
. = . . . .
.
v&α . . . BT &
θ
W&λ =1 b0T . . b0T C
[0] = [B T ]ϑ&
0 B T . . θ&.1
.
= . . . .
.
. . . . .
.
0 . . . B T θ&C
Rate of Work
W&= λ δ&
T
T
[ ]
W&= λ B 0 θ&from the kinematic law
T
W&= [λ B ]θ&
T T
0
W&= [B 0 λ ] θ&
T
W&= [b λ ] θ&
T
0
W&= λ b 0 ϑ&
T
&= M T θ& ,
D θ&* ≥ 0
P *
θ&*+1
.
.
&+C
[
&= M +1 . . M +C
D M P−1 . . . M P−C ]θ * , θ&* ≥ 0
P P θ&−1
*
.
.
θ&*−C
218
Structural Engineering Analysis and Design
Second Edition
The rate of energy dissipation terms are always positive. Also, only one of the angle terms (of the +ve and –ve
superscripted angle terms) for each particular critical section should be included in the rate of energy computation.
Flow Rule
θ&= Nθ&* , θ&* ≥ 0
θ&*+1
.
θ& 1
1 −1 .
−1 &+C
. = 1 θ * , θ&* ≥ 0
. 1 −1 θ&*−1
&
θ C 1 − 1 .
.
θ&*−C
For collapse,
W&= D&
λ b T0 ϑ&= M P T θ&* , θ&* ≥ 0
but θ&= Nθ&* , θ&* ≥ 0
λ b T0 Nθ&* = M P T θ&* , θ&* ≥ 0
b Nθ& = 1 (normalisation condition)
T
0 *
∴ λ = M P θ&* , θ&* ≥ 0
T
Hence, the mesh kinematic linear program for plastic limit analysis is
minimise λ = M T θ& P *
b T
− b & 1 (1 equation for normalisation condition, 2C terms)
T
θ * =
0 0
subject to
−B 0 (α equations for mechanism compability, 2C terms)
T T
B
positivity condition θ& ≥ 0
*
219
Structural Engineering Analysis and Design
Second Edition
Rewriting, the mesh kinematic linear program for plastic limit analysis is
θ&*+1
.
.
+C
minimise [
λ = M +P1 . . M +PC M −P1 . . . M −PC θ&*
]
θ&−1
*
.
.
θ&*−C
θ&*+1
T C
. (1 equation for the
bT . . b T0
C
− b T0 . . − b0 1
0T . normalisation condition
− BT − B C &+ C 0
T T
B . . BC . .
. θ * with 2C terms, and
subject to
. . . . . . . &−1 = 0
θ* α equations for
. . . . . . . . 0
T .
T mechanism compability
T
− Bα
T
− B α ,C
B α . . B α,C . . 0 with 2C terms)
.
θ&*−C
+ −
positivity condition θ&* ≥ 0, θ&* ≥ 0
220
Structural Engineering Analysis and Design
Second Edition
221
Structural Engineering Analysis and Design
Second Edition
Given simple frame with beam 2MP and columns MP subject to vertical 4λ and horizontal 1λ loading as shown.
The plastic collapse load factor is required.
Select critical sections C (shown by the blackened circles), identify the degree of static indeterminacy α, and
establish the sign convention for positive bending moments and rotations. For bending moments, tension is positive
and for rotations, extension is positive.
4λ
2 3 4
λ
α=2
L
C=4
1
2L
Draw bending diagrams (employing sign convention) on released structure for releases p1, p2 and external loads
defined by single parameter loading λ. We only need a load-equilibrating BMD (that not necessarily satisfies
compatibility) for the external load whilst the BMD for the releases should satisfy compatibility. The flexural
analysis for the releases p1 and p2 does not include the external loads, whilst that of λ naturally does. One solution
would be to release two rotational freedoms at 1 and 3 such that a three-pinned-arch is formed. The releases of an
internal action would necessitate the use of a bi-external action to define the p-action diagram. For instance, the
releases of an internal bending moment would require the use of a bi-external moment to define the p-BMD.
Formulate the equilibrium law equations. Leave the b0 matrix vector numerical, hence placing the variables λ, L etc
in the RHS release-loading vector.
4λ
0 -0.5 P 2
-1.5 0 -2.5
1 1 λ
0.5
1
1 0 0
P1
P1=1 diagram P2=1 diagram λ=1/L diagram
m1 1
1 0 0
3
p m 1 − p1
m C = [B b 0 ] α
2 = 2 2 p
λ m3 0 1 0 2
− 1 1 − 5 λL
m4 2
2
Formulate the mesh kinematic linear program for the plastic limit analysis of the frame. Leave MPT numerical. All
MPT values are positive and are repeated in the latter half of the row vector.
b T − b T0 & 1
minimise λ = M P θ&* 2C subject to 0T θ =
T
T *
B − B 0
positivity condition θ& ≥ 0
*
222
Structural Engineering Analysis and Design
Second Edition
θ&*+1 θ&*+1
&+ 2 &+ 2
θ* 3 5 3
θ
5 +*3
θ&+3 . 0 − 2 0 − 2 0 0 θ& .
&*+ 4 2 2 *+ 4 1
θ 1 θ&*
minimise z = [1 1 2 1 1 1 2 1] *−1 subject to
1 1 1
1 0 − −1 − 0 = 0
θ& 2 2 2 2 θ&*−1
−* 2 0 1 1 1 0 −1 − 1 − 1 &−2 0
θ&* θ*
&−3 &−3
θ* . θ* .
θ&*− 4 θ&*−4
M
positivity condition θ&* ≥ 0 and note that λ MIN = z MIN P since M TP and b 0 were left numerical.
L
Alternatively, numerical values could have been used for MP and L, in which case λMIN = zMIN. Set up and solve the
linear programme (LP) in a Simplex Solver.
PLASTIC MOMENT OF RESISTANCE, MP
1 1 2 1 1 1 2 1
KINEMATIC MATRIX
0 -1.5 0 -2.5 0 1.5 0 2.5
1 0.5 0 -0.5 -1 -0.5 0 0.5
0 1 1 1 0 -1 -1 -1
Cells to change Target cell
VARIABLES, θ to minimize
0 0 0.4 0 0.2 0 0 0.4
The changed cells in red represent the solution. The Sensitivity Report will also be produced by the solver.
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$A$10 VARIABLES, θ 0 2 1 1E+30 2
$B$10 0 1.6 1 1E+30 1.6
$C$10 0.4 0 2 4 1
$D$10 0 2 1 1E+30 2
223
Structural Engineering Analysis and Design
Second Edition
Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$J$16 SUM 1 1.4 1 1E+30 1
$J$18 SUM 0 2 0 1E+30 0.4
$J$17 SUM 0 -1 0 0.2 1E+30
Constraints
Cell Name Cell Value Formula Status Slack
$J$16 SUM 1$J$16=1 Binding 0
$J$18 SUM 0$J$18=0 Not Binding 0
$J$17 SUM 0$J$17=0 Binding 0
(a) The collapse load factor λMIN and plastic hinge angular velocities θ at collapse
λMIN = zMIN MP / L = 1.4 MP / L (take MP = 1 and λ = 1 if they have been included numerically)
θ∗+1 θ∗+2 θ∗+3 θ∗+4 θ∗−1 θ∗−2 θ∗−3 θ∗−4
0 0 0.4 0 0.2 0 0 0.4
(b) Current basic variables, potential basic variables (but as of the final tableau still non-basic) and non-basic
variables recognition
Current basic variables (participating or non-participating plastic hinges) could be zero (non-participating) as
well as have a non-zero (participating) value. Hence, current basic variables (i.e. variables in the basis in the
final tableau) are recognized by the fact that they: -
(i) have finite (zero inclusive) allowable increase in cost coefficient MPj
Note that current basic variables can have zero or non-zero final values. Current basic variables will definitely
have zero reduced cost values.
Non-basic variables include all variables that are not basic variables. Non-basic variables are recognized by the
fact that they: -
(i) have infinite allowable increase in cost coefficient MPj,
Note that non-basic variables will definitely have zero final values. Non-basic variables can have reduced cost
values of zero or non-zero.
Potential basic variables (but as of the final tableau still non-basic) are recognized by the fact that they: -
(i) are non-basic variables, hence contain all recognition conditions of non-basic variables as above, AND
(ii) have zero reduced cost
224
Structural Engineering Analysis and Design
Second Edition
225
Structural Engineering Analysis and Design
Second Edition
will be on the safe side and can be accepted for design. Hence we could just use the shadow price values (of λ,
p1, p2 … pα) off Excel in the equilibrium equations for a BMD.
m1 11 0 0 1 0 0 − 1.0M P
m
3 p1 1 3
1 − 1 − − 1M P − 0.6M
2 = 2 2 p = 2 2 2M = P
m 3 0 1 0 2 0 1 0 P
2.0M P
1 5 λ L 1 5 MP
m 4 − 2 1 − − 1 − 1.4 L L − 1.0M P
2 2 2
Check that the bending moment everywhere is less than the particular plastic moment of resistance of the member.
If so, we have correctly chosen the critical sections C that will produce the smallest value of λ. And hence the Yield
Condition is met and λ is the unique collapse load factor according to the Uniqueness Theorem.
To reiterate the above results for ease of understanding, we make the following observations in this order: -
I. The critical plastic load factor is λP = 1.4 MP / L shown in the target cell to minimize and again in the
Shadow Price of the Sensitivity Report.
II. Identify the basic variables, i.e. the locations that a plastic hinge forms to form the collapse mechanism.
This is identified in the Sensitivity Report on variables with a Finite (zero included) Allowable
Increase. Here, they are θ3, θ1 and θ4, the latter two of which are negative. This means that they rotate
such as to cause compression on the positive sign convention side. Usually (but not necessarily) the basic
variables will be non-zero.
4λ
2 3 4
λ
The value of the Allowable Increase is the amount by which the objective function coefficient i.e. the MP
value corresponding to the basic variable can increase before the solution (i.e. values of the angular
velocity θj) changes. This is useful to force another (maybe less catastrophic) collapse mechanism. To
do this the structural engineer would have to increase the MP value at section 3 by 4MP, or at section 1 by
0.5MP or at section 4 by 0.67MP. Clearly, if say the engineer does not want the column to collapse
(that being catastrophic), he would have to increase the MP of the column at section 1 to a little
over 1.5MP. If that were done, in this case, sections 2, 3, and 4 would then form the mechanism, i.e. a
less catastrophic beam mechanism.
III. Next, we look for the existence of an alternate collapse mechanism. For each variable θj, the reduced
cost gives the increase in the target cell (z) for a unit change in the variable θj. Thus the reduced cost of
all the basic variables, i.e. the plastic hinges must be zero for the optimal conditions (i.e. minimum λ).
Thus, if any reduced cost is zero for a non-basic variable (infinite Allowable Increase), there can be an
alternate collapse mechanism having the same load factor because this non-basic variable could be
brought into the basis to form another collapse mechanism without increasing the plastic collapse load
factor. Here there is not any alternate collapse mechanism since all the variables with infinite
Allowable Increase have non-zero Reduced Costs.
226
Structural Engineering Analysis and Design
Second Edition
m1 11 0 0 1 0 0 − 1.0M P
m
3 p1 1 3
1 − 1 − − 1M P − 0.6M
2 = 2 2 p = 2 2 2M = P
m 3 0 1 0 2 0 1 0 P
2.0M P
1 5 λ L 1 5 MP
m 4 − 2 1 − − 1 − 1.4 L L − 1.0M P
2 2 2
Check that the bending moment everywhere is less than the particular plastic moment of resistance of the
member. If so, we have correctly chosen the critical sections C that will produce the smallest value of λ.
And hence the Yield Condition is met and λ is the unique collapse load factor according to the
Uniqueness Theorem. If the yield condition were not met, we would have reduced the estimate of critical
plastic collapse factor as follows.
M Exceeding
λ Pr educed = λ P
MP
And we would say that the true critical collapse load factor lies between
λ Pr educed < λ Ptrue < λ P
227
Structural Engineering Analysis and Design
Second Edition
Given MP = 590kNm
228
Structural Engineering Analysis and Design
Second Edition
229
Structural Engineering Analysis and Design
Second Edition
230
Structural Engineering Analysis and Design
Second Edition
3.3.5 Plastic Limit Analysis of Rectangular Multi-Storey Multi-Bay Frames with Improved Data
Generation
(a) Select critical sections inclusive of articulation C*, identify the degree of static indeterminacy with articulation
ignored α∗, and establish the sign convention for positive bending moments and rotations
Critical sections inclusive of articulation C* are identified.
α for plane frames with articulation = 3 (number of windows or cells) – releases r
α∗ for plane frames with articulation ignored = 3 (number of windows or cells)
Bending moments and rotations are defined as positive if they cause tension or extension at the side with the
dotted lines. For multi-storey multi-bay buildings, the dotted line is drawn under beams and on the right of
columns by convention.
p
m C * = [B b 0 ] α *
λ
(i) 3 self equilibrating BMDs (due to shear, bending and axial actions) for each cell (sign convention applies)
Note that C* is the number of critical sections inclusive of those originally articulated and α∗ is the degree
of static indeterminacy with all articulation ignored. The articulation positions will thus also be regarded
and denoted as critical sections. Hence, each cell will contribute 3 indeterminate forces and hence 3
columns in the B matrix. Fill in the B matrix column-by-column with the bending moment values of the 3
normalized cell self-equilibrating BMDs at the critical section positions of C*.
(ii) 1 load equilibrating BMD for the load factor λ (sign convention applies)
Vertical loads are transferred through simply supported beams, and hence MMAX = PL/4 if central load.
Horizontal loads are transferred through cantilever action and multiple loads on the same column have their
independent BMDs superimposed upon each other.
(c) Formulate the mesh kinematic linear program for the plastic limit analysis of the frame and solve the LP
Set the MP value of the real articulation (hinges) to zero in the MPT matrix.
b T − b T0 & 1
minimise λ = M P θ&* 2 C* subject to T0 θ =
T
T *
B − B 0
positivity condition θ&* ≥ 0
231
Structural Engineering Analysis and Design
Second Edition
232
Structural Engineering Analysis and Design
Second Edition
233
Structural Engineering Analysis and Design
Second Edition
An unbraced and rigid-jointed frame fabricated from S275 steel with pinned bases is shown. An investigation of the
possibility of elasto-plastic instability is to be carried out to establish what load factor, applied to the given load
would be required to cause such a failure.
234
Structural Engineering Analysis and Design
Second Edition
235
Structural Engineering Analysis and Design
Second Edition
A unique bending moment distribution is obtainable if and only if there are α*+1 active participating hinges.
Here, α*+1 = 10, and the number of active participating hinges is 6 from above.
Hence, the exact bending moment distribution at collapse is not unique and is indeterminate.
236
Structural Engineering Analysis and Design
Second Edition
237
Structural Engineering Analysis and Design
Second Edition
238
Structural Engineering Analysis and Design
Second Edition
239
Structural Engineering Analysis and Design
Second Edition
The linear programming simplex method described above performs the plastic collapse analysis meeting the
fundamental theoretical requirements as follows
(i) The equilibrium equation is employed, hence satisfying the equilibrium condition
(ii) Many different collapse mechanisms is postulated, hence satisfying the mechanism condition, and a linear
solving algorithm is used to find the smallest λ
(iii) The yield condition is automatically met since λ is minimized assuming that the chosen critical sections
contains the critical mechanism.
The hand method of plastic collapse analysis for simple structures of low degrees of static indeterminacies and a
small number of critical sections is as follows
(i) The equilibrium equation is employed, hence satisfying the equilibrium condition
(ii) A collapse mechanism is postulated i.e. values and signs of the assumed bending moments MP is put into
the bending moment at collapse {m} vector, hence satisfying the mechanism condition, and λ is solved for.
(iii) Then {m} is computed and it is ensured that at no point of the structure, the bending moment is greater
than the local MP value, hence satisfying the yield condition. Hence, naturally the smallest λ is found.
Given simple frame with beam 2MP and columns MP subject to vertical 4λ and horizontal 1λ loading as shown.
The plastic collapse load factor is required.
4λ
2 3 4
λ
α=2
L
C=4
1
2L
Appreciate that to form a mechanism we need α + 1 plastic hinges. Postulate Mechanism Condition with hinges at
1, 3 and 4.
4λ
2 3 4
λ
Formulate the equilibrium conditions noting the sign convention for positive bending moments and rotations. For
bending moments, tension is positive and for rotations, extension is positive. 4λ
0 -0.5 P2
-1.5 0 -2.5
1 1 λ
0.5
1
1 0 0
P1
P1=1 diagram P2=1 diagram λ=1/L diagram
240
Structural Engineering Analysis and Design
Second Edition
− M P 1
1 0 0
3
p M 1 − p1
m C = [B b 0 ] α 2 = 2 2 p
λ 2M P 0 1 0 2
1 5 λ L
− M P
− 1 −
2 2
Sign convention for
extension positive Sign convention for
tension positive
Note that we only need a load-equilibrating BMD (that not necessarily satisfies compatibility) for the external load
whilst the BMD for the releases should satisfy compatibility. Let us try a different load-equilibrating diagram that
DOES NOT satisfy compatibility, say pinned beams and cantilever columns. Postulate the same mechanism.
4λ
MP 1 0 − 1 0 0
M 0.5 p1
2 = 1 0 λ
p2
2M P 0 1 2 2
λL
− M P − 0.5 1 0 1
Solving these equations simultaneously will again result in λ = 1.4MP/L. But now, p1 = 0.4MP and p2 = −0.8MP.
The values of p1 and p2 are no longer physically meaningful. Before it used to be the values of the indeterminate
forces, now it is meaningless because of the non-compatible load-equilibrating BMD.
241
Structural Engineering Analysis and Design
Second Edition
242
Structural Engineering Analysis and Design
Second Edition
243
Structural Engineering Analysis and Design
Second Edition
244
Structural Engineering Analysis and Design
Second Edition
245
Structural Engineering Analysis and Design
Second Edition
246
Structural Engineering Analysis and Design
Second Edition
247
Structural Engineering Analysis and Design
Second Edition
248
Structural Engineering Analysis and Design
Second Edition
249
Structural Engineering Analysis and Design
Second Edition
250
Structural Engineering Analysis and Design
Second Edition
251
Structural Engineering Analysis and Design
Second Edition
252
Structural Engineering Analysis and Design
Second Edition
253
Structural Engineering Analysis and Design
Second Edition
254
Structural Engineering Analysis and Design
Second Edition
255
Structural Engineering Analysis and Design
Second Edition
256
Structural Engineering Analysis and Design
Second Edition
257
Structural Engineering Analysis and Design
Second Edition
258
Structural Engineering Analysis and Design
Second Edition
259
Structural Engineering Analysis and Design
Second Edition
260
Structural Engineering Analysis and Design
Second Edition
261
Structural Engineering Analysis and Design
Second Edition
3.4 GNL, MNL, Contact Nonlinear Static and Buckling Analysis by The Implicit Tracing The
Equilibrium Path Method
An equilibrium path that does not include the initial unloaded state (λ=0,
Secondary equilibrium path
{U}={0})
An equilibrium path for which the displacements are zero for all levels of
Trivial fundamental path
loading i.e. (λ≠0,{U}={0})
262
Structural Engineering Analysis and Design
Second Edition
The static equilibrium equation has been established, and is restated in all its simplicity.
[K T ]δ{U} = δ{P}
where [K T ] = [K E ] + [K G ]
where [K E ] = [T ] [[k E ]][T ]
T
∂2 d
and [K G ] = [T]T [[k G ]][T] + {f }
∂{U}∂ U
∂ ε2
∂ 2 Wi ∂ 2 Wn
for which [k G ] = ∫ {{σ}n + {σ}i − [ D]{ε}i }dΩ − − λ
Ω
∂{d}∂ d ∂{d}∂ d ∂{d}∂ d
and {σ}n = [D][B]{d}
and [
{f } = ∫Ω [B]T [D][B]dΩ {d} + ] {∫ [B] {σ} dΩ}− {∫ [B] [D]{ε} dΩ}− {∫ [N] {b}dΩ}
Ω
T
i
Ω
T
i
Ω
Τ
In order to find the buckling load, we could in theory perform a nonlinear eigenvalue solution. The nonlinear
eigenvalue problem is expressed as
[KT]{φ}={0}
det | [KT] | = 0
This is quite a complex solution scheme is thus is not usually attempted. As closed-form mathematical solutions of
realistic multi-degree of freedom nonlinear eigenvalue problems are highly complex, incremental methods to
establish the equilibrium path is used. The incremental procedures can correspond to an increment of displacement
or an increment of loading, i.e. displacement control or load control. The incremental approach is normally started
from the initial unloaded equilibrium state, in which case the fundamental path is traced. In the vicinity of
bifurcation states, it is essential that small incremental steps be used so as to avoid a jump from the fundamental
path to a different path.
263
Structural Engineering Analysis and Design
Second Edition
λi
{G}j ESi
Full Newton
∆λ Raphson
{P}j
{R}j
λi-1 ESi-1
U
Ui-1 ∆U Ui
The vector {f} includes the fixed end forces which are negative terms here. It is equivalent then to add those terms
to the {P} vector and call it the equivalent nodal loading vector if we desire; however the mathematics will work
out nevertheless as the fixed end forces are included within {f} as negative terms. Now, in order to trace the
nonlinear load path, we formulate a set of linear equations valid over a small load or displacement step.
[KT]δ{U} = δ{P}
This is the linearized equilibrium equation valid as long as [KT] is approximately constant over the displacement or
load step. We can also rewrite the above equation as
[KT]δ{U} = −{G}
where {G} is the out-of-balance force between the equivalent loads and the resistance forces. In this way, there will
be no accumulation of error as the load steps are traced.
264
Structural Engineering Analysis and Design
Second Edition
The modified Newton-Raphson is efficient when there is no strain hardening or an unloading path is not required,
otherwise the Newton-Raphson is required for convergence.
We start with load control. Load control used for ascending portions of the equilibrium path. When close to a
bifurcation point, load control is used with small steps. Then when close to a limit point, displacement control is
used. Displacement control is also used for descending parts of the equilibrium path. Then, if we come to a limit
point in the displacement sense, we revert to load control or use another displacement parameter.
Advancing Phase
1. Determine an increment (of load, displacement or arc length) to trace on the equilibrium path from ESi-1 to ESi
2. Determine an estimate of the tangent stiffness at ESi-1, KTi-1
3. Determine the predicted displacement increment to move forward by solving the equilibrium equations
[KTi-1]∆{U}Pi-1 = −{G}i-1
4. Calculate the element resisting force {R}i-1
5. Calculate the unbalanced force {G}i-1 and check for convergence. If converged, i.e. {G} ≈ 0, then go to 1 with i =
i +1. If not converged go to 6.
Note that no convergence can be achieved if the total applied load is greater than the buckling load. The Full
Newton-Raphson procedure can be expensive and unnecessary when the solution is close to convergence.
265
Structural Engineering Analysis and Design
Second Edition
3.4.3 Equilibrium Paths, Stability of Equilibrium Paths, Critical Points, Stability of Critical Points
Apart from the linear eigenvalue method (linear elastic buckling) and nonlinear tracing the equilibrium path,
stability if an elastic system can be interpreted by means of the concept of minimum potential energy. An elastic
system always tend to go to a state in which the total potential energy is at a minimum. The types of instability
include: -
Bifurcation instability can be analyzed using linear buckling analysis whilst limit load instability is analyzed by
tracing the equilibrium path.
Load
Deflection
266
Structural Engineering Analysis and Design
Second Edition
separately to yield two values defining the bifurcation point. We can also identify the limits points
for imperfect systems by differentiating the equilibrium path equation, i.e. dP/dθ = 0. Hence we
shall obtain value for θ at the limit point, θlp which subsequently is replaced into P for the value of
the critical point, Plp.
(v) Investigate the stability of the critical point. The bifurcation point (θ = 0, P = Pcr) or the limit point
(θ = θlp, P = Plp) is replaced into the third derivative of V.
≠0 ∂ 3V
Unstable
∂θ 3
=0
<0 ∂ 4V >0
Unstable Stable
∂θ 4
=0
=0
Neutral stability
An equilibrium state is stable if ∆V is positive for all possible infinitesimal disturbances δ{U}. This implies that an
agent external to the system will have to perform positive work in order to cause δ{U}.
267
Structural Engineering Analysis and Design
Second Edition
Critically stable or Gaussian elimination leads to at least one zero diagonal term,
δ2V positive
unstable ES i.e. δ2V = 0.5 δ{U}T[KT]δ{U} = 0 for some δ{U} = δ{U0}
semi-definite
i.e. a critical point or det ([KT]) = 0
A critical point (buckling point) can be a bifurcation or a limit point. At a critical point det ([KT]) = 0, i.e. KT is
singular indicating the presence of an infinitesimally adjacent equilibrium state at the same level of loading. A
critical point is normally associated with one buckling mode δ{U0}. Bifurcation points are very rare in real
structures due to the presence of imperfections in the geometry and due to pre-buckling displacements.
A critical point is a limit point if and only if the first order work
Limit point done by {Pn} over the buckling mode(s) δ{U0} is non-zero. {Pn}Tδ{U0} ≠ 0
Loading cannot be increased further.
268
Structural Engineering Analysis and Design
Second Edition
3.4.4.1 GNL, MNL Load Control, Displacement Control or Arc-Length Control Static Analysis
All cards applicable to SOL 101 are also applicable to SOL 106.
SOL 106
$ CASE CONTROL SECTION
Unlike linear analyses schemes, nonlinear analysis schemes (SOL 106 and SOL 129) employ subcases on an
incremental basis, instead of separate load cases and boundary conditions. In linear analysis, subcases represent an
independent loading condition. Each subcase is distinct from each other. Nonlinear static analysis permits only one
independent loading condition per run. For SOL 106, the bulk data loads and prescribed displacements are
measured from the initial configuration. Hence latter subcases should refer to different load cards that have
magnitudes that include the magnitude of load cards from previous subcases. Hence, if the same point is being
loaded or displaced in a subsequent subcase, the load or displacement value referred by the subsequent subcase
should be relative to the initial undeflected value of zero and not that of the previous subcase. Subcases are hence
269
Structural Engineering Analysis and Design
Second Edition
cumulative. Loads and boundary conditions at the end of a subcase are the initial conditions for the next subcase.
Hence if the load description is inadvertently omitted in subsequent subcases, the load will automatically be
removed in an incremental fashion. Only one independent loading history can be applied during an analysis. In
general, a different loading sequence requires a complete new analysis. The load to be applied during a particular
stage (i.e. subcase) is referred to by the LOAD Case Control Command. NINC in the NLPARM bulk data entry is
an integer that specifies the number of increments in the particular subcase, by default 10. The more linear the
problem is, the smaller the value of NINC that can be acceptable. But when nonlinearity is significant such as at the
onset of plasticity, at regions of stress concentration where each load step causes high stress changes, at contact
regions with high forces and at the onset of buckling, the incremental load should be small (i.e. NINC must be
large). Hence different subcases should be used to model different regions along the load path. Where nonlinearity
of stiffness is small, NINC can be small, and when nonlinearity is large, NINC must be large. The load specified in
a particular subcase minus the load specified in the preceding subcase is equally divided by NINC to obtain the
incremental load for the particular subcase. Clearly, if convergence is a problem, this incremental load should be
made smaller (by increasing NINC and/or decreasing the magnitude of load specified in the particular subcase
relative to the previous subcase) until convergence. Self-weight and/or prestressed elements must be simulated by
incorporating an initial load case followed by whatever static loading load case. Each subcase should not have
NINC exceeding 1000 to control the database size, output size and restarts.
The follower force is calculated and incorporated by the use of PARAM, FOLLOWK, YES. We know how the
prestress affects the differential stiffness, namely a tensile prestress causing an increase in stiffness. The effect of the
follower force on the stiffness is different. For example, for a cylinder under external pressure critical buckling load may
be over-estimated (even though the mode shapes are similar) in a SOL 105 and the natural frequencies in vibration may
be under-estimated (even though the mode shapes are similar) in a SOL 103 in the absence of follower stiffness. To the
contrary, these observations are reversed in case of centrifugal loads. Centrifugal forces as a constant (static) load are
applied by a Bulk Data RFORCE to any elements that have masses. The follower stiffness due to centrifugal load has the
effect of lowering stiffness (although the centrifugal load tensioning effect increases stiffness), consequently lowering
natural frequencies (even though the mode shapes are similar) in a SOL 103 and lowering the buckling loads (even
though the mode shapes are similar) in a SOL 105. This effect increases as the RPM increases, and it becomes significant
when the RPM is over 1000. For moderately geometric nonlinear analysis, exclusion of follower stiffness affects the rate
of convergence, but the converged solution is correct. For severely geometric nonlinear analysis, it may not be
possible to obtain a converged solution without including follower stiffness. As the geometric nonlinearity
intensifies, so is the effect of follower stiffness. Therefore, inclusion of follower stiffness greatly enhances the
convergence if the deformation involves severe geometric nonlinearity.
Output requests for each subcase are processed independently, but are appended after computation for output
purposes. The INTOUT field is specified with YES, NO or ALL and determines whether the output is requested
for each load increment within a subcase or just the last load step of the subcases.
YES Output requested for every computed converged load increment within the subcases
NO Output requested for only the last converged load increment of the subcases
ALL Output requested for every computed and user specified load increment within the subcase
INTOUT is especially important for restarts. For Newton’s iteration methods (i.e. without NLPCI) the option ALL
is equivalent to YES but not when arc length methods (with NLPCI) are used.
Load control is specified by having load card definitions. Naturally, the LOAD case control command is required.
Load cards relevant to SOL 101 is also relevant to SOL 106 except for DEFORM cards. The applicable load cards
are FORCEi, MOMENTi, PLOADi. GRAV, TEMP, SPC, SPCD etc. Loads which are stationary in direction
throughout the analysis include: -
270
Structural Engineering Analysis and Design
Second Edition
Displacement control is specified by having SPCD card definitions, noting that grid points with displacements
enforced by SPCD must also have an SPC entry. Note that enforced displacement can also be specified directly on
the SPC bulk data entry, although this method is not recommended. Naturally, the LOAD case control command is
required.
Arc-length control is specified by having load card definitions and also a NLPCI card with the same ID as the
NLPARM card. Arc-length control cannot be used in conjunction with displacement control SPCD cards. The arc-
length method is useful to trace unstable or post-buckling equilibrium paths defined by negative load control steps.
The critical buckling load can be estimated by Newton’s load control method by performing the nonlinear static
analysis until the solution cannot be obtained due to divergence, in which case the adaptive bisection method is
activated in the vicinity of the critical buckling load and stops at the limit load close to the critical buckling load.
But beyond that, the load control method does not work, and hence the arc-length method is required.
MAXR on the NLPARM card is the maximum ratio for the adjusted arc-length increment relative to the initial
value, by default 20.0. The TYPE field on the NLPCI card could be CRIS (default), RIKS or MRIKS.
The load is divided up into load steps. Each load step is then divided into iteration steps. KMETHOD specifies the
method for controlling stiffness updates, whether AUTO (default), SEMI or ITER.
With AUTO, NASTRAN automatically selects the most efficient strategy of when to update the tangent stiffness
matrix based on convergence rates. The matrix could be updated in the middle of an iteration and will not be
necessarily be updated at the beginning of the load step. AUTO is a good starting method. It essentially examines
the solution convergence rate and uses the quasi-Newton, line search and/or bisection convergence acceleration
methods to perform the solution as efficiently as possible without a stiffness matrix update. Highly nonlinear
behaviour in some cases may not be handled effectively using AUTO. For AUTO (like SEMI), the stiffness matrix
is updated on convergence if KSTEP is less than the number of iterations that was required for convergence with
the current stiffness.
SEMI is similar to AUTO (it also uses the quasi-Newton, line search and/or bisection convergence acceleration
methods) except that the tangent stiffness matrix will also be updated at the beginning of each load step,
irrespective of the convergence status. Hence SEMI provides better convergence at the expense of higher cost. The
SEMI method is an efficient method for nonlinear static iteration. For SEMI (like AUTO), the stiffness matrix is
updated on convergence if KSTEP is less than the number of iterations that was required for convergence with the
current stiffness.
With ITER, the stiffness matrix is updated at every KSTEP iteration. KSTEP specifies the number of iterations
before the stiffness update, 5 by default. It is a ‘brute force’ method. It can be most suited for highly nonlinear
problems. Full Newton-Raphson is obtained by selecting KMETHOD = ITER and KSTEP = 1. Modified Newton-
Raphson is obtained by selecting KMETHOD = ITER and KSTEP = MAXITER.
271
Structural Engineering Analysis and Design
Second Edition
3.4.4.1.3 Convergence Acceleration Techniques (Quasi-Newton MAXQN, Line Search using MAXLS and
LSTOL, Bisection using MAXITER, RTOLB, MAXBIS and MAXDIV)
The convergence acceleration techniques include quasi-Newton, line search and bisection.
The stiffness updating schemes can be supplemented with quasi-Newton (QN) updates. MAXQN is the maximum
number of quasi-Newton correction vectors to be saved on the database, equal to MAXITER by default. The BFGS
update is performed if MAXQN > 0.
The line search is controlled by MAXLS and LSTOL. MAXLS is the maximum number of linear searches
allowed for each iteration, default 4. LSTOL requires a real number between 0.01 and 0.9 to specify the tolerance
for the line search operation, default 0.5. The line search operation will be conducted if the error defining the
divergence rate is greater than LSTOL. Computing cost for each line search is comparable to that of an iteration.
MAXITER specifies the limit on number of iterations for each load increment, 25 by default. If the solution does
not converge at MAXITER iterations, the load increment is bisected and the analysis is repeated. Bisection is also
activated if RTOLB, by default 20.0 is exceeded. RTOLB is the maximum value of incremental rotation (in
degrees) allowed per iteration before bisection is activated. MAXBIS specifies the maximum number of bisections
for each load increment, 5 by default. If MAXBIS is positive, the stiffness matrix is updated on the first divergence
and the load increment is bisected on the second divergence. If MAXBIS is negative, the load increment is bisected
every time the solution diverges until the limit on bisection is reached. If |MAXBIS| has been attained and the
solution still has not converged then the value on MAXDIV, which specifies the divergence criteria for an
iteration, 3 by default, is considered. If MAXDIV is positive, the best attainable solution is computed and the
analysis is continued to next load increment. If MAXDIV is negative, the analysis is terminated. Hence, if we want
the analysis to terminate if convergence has not been achieved, MAXDIV must be negative, not the default value.
The recommended value for MAXDIV is thus −3.
The convergence test is performed at every iteration. CONV specifies the flags to select convergence criteria, UPW
or any combination, note default PW. EPSU is the error tolerance for displacement error (U) convergence criteria.
EPSP is the error tolerance for load equilibrium error (P) convergence criteria. EPSW is the error tolerance for
work error (W) convergence criteria. If displacement control is used, then there is a need for tighter tolerances for
convergence. There is a facility to provide a default value depending on the type of analysis, type of elements and
desired accuracy by specifying the NLTOL parameter and leaving the CONV, EPSU, EPSP and EPSW fields in the
NLPARM blank. Models with gap, contact or heat transfer elements will have tighter default convergences.
$ BULK DATA
272
Structural Engineering Analysis and Design
Second Edition
(Default)
EPSP=1.0E-2
1 High
EPSW=1.0E-3
EPSP=1.0E-2 EPSP=1.0E-3
2 Engineering EPSW=1.0E-2 EPSW=1.0E-5
(Default) (Default)
EPSP=1.0E-1
3 Preliminary
EPSW=1.0E-1
The quality of the mesh is essential in nonlinear analysis. Artificial stress concentrations due to poor modelling
such as when beams interface with a shell mesh without special handling techniques will cause convergence
difficulty. The mesh should also be adequately fine for good stress recovery, hence a h-element convergence
exercise must be undertaken. This could probably be performed by repetitive linear analysis, the resulting
converged mesh can be used with confidence in a nonlinear analysis. Other steps that can be taken to improve
convergence are increase the number of load increments, break applied load into multiple subcases, use a more
general material stress-strain definition instead of a bilinear model, avoid poorly shaped elements in regions of high
stresses and introduce initial load steps in the solution to establish equilibrium in the model prior to increasing the
nonlinearity. Contact regions should have a fine mesh in order to capture the contact stresses. If large deformations
are expected to distort elements such that their accuracy may be called into question, it may be worthwhile to
manually distort the elements in the opposite direction somewhat before starting the solution so that the final shape
will be closer to the ideal shape.
Nonlinearity
Nonlinearity
Work Done
Differential
With {d}
Stiffness
Material
Element
Mass
CBAR
PBAR Yes No Yes No Yes No No
MAT1 (MATT1)
CBUSH
No Yes Yes No No No Yes
PBUSH and PBUSHT
CBUSH1D
Yes Yes Yes Yes Yes No Yes
PBUSH1D
CROD
PROD Yes No Yes Yes Yes No Yes
MAT1 (MATT1), MATS1
CONROD
Yes No Yes Yes Yes No Yes
MAT1 (MATT1)
CTUBE
PTUBE Yes No Yes Yes Yes No Yes
MAT1 (MATT1)
CBEAM
PBEAM Yes No Yes Yes Yes No Yes
MAT1 (MATT1), MATS1
CQUAD4, CTRIA3 Yes No Yes Yes Yes No Yes
273
Structural Engineering Analysis and Design
Second Edition
PSHELL
MAT1 (MATT1), MAT2
(MATT2), MAT8, MATS1,
CREEP
CQUAD8, CTRIA6
PSHELL Yes No Yes No Yes No No
MAT1 (MATT1), MAT2
CQUAD4, CTRIA3
PLPLANE Yes No Yes Yes Yes Yes Yes
MATHP
CQUAD8, CTRIA6
PLPLANE Yes No Yes No Yes Yes No
MATHP
CHEXA, CPENTA, CTETRA
PSOLID
Yes No Yes Yes Yes No Yes
MAT1 (MATT1), MAT9
(MATT9), MATS1, CREEP
CHEXA, CPENTA, CTETRA
PLSOLID Yes No Yes Yes Yes Yes Yes
MATHP
Note that the CBAR element is not a nonlinear element i.e. its coordinates are not updated as the element
deforms.
The CBUSH is a generalized nonlinear spring - linear damper element. The CBUSH1D is a rod-type nonlinear
spring - nonlinear damper element; it is a 1D version of the CBUSH element with added features. The CBUSH is
NOT geometrically nonlinear, only materially. The CBUSH1D element can have geometric and material
nonlinearity. The CBUSH1D element is the only element which can model rotation damping such as for
modelling tuned pendulums mass dampers.
The CROD, CONROD, CTUBE elements can be used to model cables as these elements are geometrically
nonlinear, incorporates the differential stiffness and can refer to different stress-strain curves for tension and
compression.
CGAP element can also be used to work in the opposite sense to a contact element in order to function as a cable
element. That is to say, the element can be made to work only in tension by specifying a separation distance
corresponding to the taut condition, above which the element is to be turned on and resist tension.
The CBEAM element can model plastic hinges at its ends for plastic collapse analysis of frameworks. For this,
the material type specified in the MATS1 card must be elastic-plastic (and perfectly plastic too) i.e. TYPE
PLASTIC, and strictly not nonlinear elastic. The CBEAM is assumed to have its nonlinearity concentrated at its
ends. Obviously if material nonlinearity is enabled, there can be no pin flags. The user need not specify the cross-
section axis about which the yielding occurs, since the implementation allows for combinations of bending
moments in two directions plus an axial load. The flexibility of the plastic hinge is based upon eight idealized rods
at each end, chosen to match the total area, center of gravity and moments of inertia of the cross section.
Note that rigid body elements RBEi, RBAR and RROD do not rotate in geometric nonlinear analysis i.e. their
coordinates are not updated as the element deforms and hence are not nonlinear elements. Geometric stiffness is
also not computed for rigid elements. Offsets are not recommended in nonlinear analysis, or even linearized
buckling analysis. Overly stiff elements, as replacements are also not recommended as they may cause problems
with convergence (MAXRATIO exceeded). Too stiff will give you mathematical problems and too flexible will
create too soft a connection. Thankfully what you might consider "very stiff" is numerically still likely to be OK. In
number terms you want something that is maybe 1000X stiffer than the local structure. If you are still unsure try 2
274
Structural Engineering Analysis and Design
Second Edition
runs, with one having a bar 10 times stiffer than the other. If the answers are similar then you know it is stiff
enough. To estimate the stiffness of the structure at that point connecting the gap, use simple beam formulae such
as 12EI/L3 or 3EI/L3 or alternatively, apply an arbitrary load and ascertain the displacement from a SOL 101 static
analysis.
Midside grids on solid elements are not allowed for nonlinear analysis.
There are three primary types of contact elements, namely point-to-point contact gap elements (CGAP), line-to-
line or surface-to-surface slide line contact elements (BLSEG) and general arbitrary line-to-line or surface-
to-surface contact elements.
The CGAP element is a point-to-point contact element. Its element coordinate system is defined depending on
whether the end grids points are coincident or not. If coincident, the orientation is defined by using the coordinate
system CID, which if unspecified would default to the basic system. If the grid points are not coincident, the
orientation is defined using the element x-axis using defined by the GA to GB grid orientation and the v vector is
defined using X1, X2, X3 or G0 entries. Note also that no stress singularity will be introduced here by connecting
many 1D gap elements to the shell or solid mesh. This is because of the multiple 1D element. Had there been just
one 1D element connected to a shell or solid mesh, then certainly there will be stress singularity.
The two basic properties of the CGAP element are its initial gap and its compressive stiffness. If the initial node-
to-node separation becomes more positive, the gap has no stiffness and is for all intents and purposes dormant. As
the separation becomes less positive and approaches zero or some predefined initial gap, the element wakes up and
becomes a stiff spring, which resists further closure. The point at which the gap begins to resist this displacement is
defined by the initial gap. This initial gap may be zero but the length of the element may not. The length of the gaps
may be zero for solid elements by may not for shell elements, whereby the user may decide that due to mid-plane
modelling, these two plates that might rest on each other, would by definition, be separated by half the sum of their
thicknesses. By bridging this separation with gap elements assigned with initial gap of zero, the code will know that
this initial position represents the two parts already in contact.
The orientation is of the CGAP (whether coincident or not) is important to define the direction of opening and
closing as depicted in the figures below. Hence GA and GB must be attached to the correct portions of the model in
order to simulate the correct opening and closing behaviour.
The preload F0 is useful to model friction because the frictional force is a function of the total compressive force
which in turn is a function of the preload. Hence in a SOL 129 analysis, if the only contribution of gravity is to add
275
Structural Engineering Analysis and Design
Second Edition
a normal force which would result in frictional forces, the normal forces induced in the contact elements can be
specified as preloads instead of modelling the gravity load case explicitly.
Gap element stiffness controls the numerical stability. The compressive stiffness must simulate the local stiffness
of the elements in the contact pair. If the stiffness is too high, the element may bounce as the load is incremented
near zero separation. The closed gap stiffness should not exceed the adjacent DOF stiffness by 1000 times. If the
stiffness is too low, external forces may actually drive one node through the other. The CGAP element should be
used when there is very little relative displacement or sliding normal to direction of closure is expected. The
process of choosing a gap stiffness may be an iterative one as an initially soft gap if stiffened to the point where
penetration is minimal but the nonlinear solution still converges. Alternatively, the adaptive gap stiffness
technology adjusts the gap stiffness internally to achieve convergence without penetration within some predefined
tolerance. In the adaptive solution, the method of specifying the initial gap stiffness is the opposite of the manual
iterative process. Choose a stiffer gap and let the code soften it as is required numerically. The adaptive method is
clearly more efficient. Note that the mesh on either contacting surfaces should correspond exactly so that the gap
elements are well aligned. Relative slide should be insignificant as this can cause the gap element to go into tension
when in reality it is still in contact. This problem arises if the force in the gap element acts in the direction of the
element. A better method would then certainly be to define the stiffness of the gap element in a particular DOF
only i.e. normal to the contact interface.
The gap element provides point-to-point contact. When the gap element is open, there is no contact and no friction.
When the gap element is closed, there are three different conditions. The first case is when the gap is sliding (no
friction). The second case occurs when the gap element is sticking (static friction). The third case occurs when the
gap element is slipping (kinetic friction). Sometimes, it may be difficult to converge because of the switching
among these conditions. Typical coefficients of friction are presented.
On the PGAP Bulk Data entry, U0 and F0 are the initial gap opening and the preload, respectively. U0 is the
separation of the gap element. If you are unsure about what preload to use, use the default value of zero. KA, KB,
and KT are called the penalty values. KA is the axial stiffness when the gap is closed, KB is the axial stiffness
when the gap is open, and the KT is the transverse gap stiffness. All you need to input is the KA value. KB is
defaulted to be zero, and KT is defaulted as a function of KA. The selection of KA or a penalty value is critical to
the convergence of the problem. The non-adaptive gap does not update this KA value. The new adaptive gap
updates this KA value if you specify a positive value for TMAX. The default option for the new adaptive gap is NO
adaptive stiffness update. This is when the TMAX value is equal to 0.0. If TMAX is a negative number, then
MSC.Nastran uses the non-adaptive gap formulation. A good initial KA selection is roughly 1000 times the
Young's modulus of the material that is in the neighborhood of the gap element. Fields 8 and 9 also require some
276
Structural Engineering Analysis and Design
Second Edition
explanation. For the new adaptive value this is MU1, which is the static coefficient of friction, and MU2, which is
the dynamic coefficient of friction. The non-adaptive gap elements use MUY and MUZ in the same fields instead
of MU1 and MU2 values. Therefore, to compare the results using adaptive and non-adaptive gap elements, set both
of these entries to the same values. On most problems, you will not know how to estimate the penetration depth or
field 2 on the continuation entry. Therefore, if you skip the continuation entry, the default new adaptive gap is
obtained, which is everything adaptive except the KA update. The recommended allowable penetration of TMAX
is 10 percent of the thickness for the shell element, and for the solid element it is roughly 1/10,000 of the
characteristic length for the problem. A small positive value will turn on the adaptive stiffness update, which will
then converge at a faster rate. The maximum adjustment ratio or MAR in field 3 of the continuation entry is used
only for the penalty value adjustment of the adaptive gap element. The default value of 100 is sufficient for most
problems.
Further discussion of CGAP parameters is warranted. U0 is self explanatory - the initial open value of the gap. F0 -
any preload you want to apply - be careful if KB is the only thing preventing rigid body motion. KA - this is the
source of most convergence problems due to gap elements. It can be calculated from the local stiffness of the
structure at the GRIDs to which the gap are applied. If you can't do this by hand (usually the case), then run a linear
static (SOL 101) analysis with U0 temporarily set greater than 0.0 (or remove the gap element from the model for
this run, if possible); this is so the gap stiffness you have already defined at KA does not influence the local
stiffness. Then apply a unit load to both GRIDs of the gap in the direction of the gap X axis. Run the analysis and
recover the displacement of both GRIDs in the gap X direction. Now take 1000 times the inverse of the
displacement at the gap GRIDs and use the smallest value of the two for the value of KA. Now 2 questions arise.
First, if the value of displacement in the gap X direction recovered from this run is, say, 3.4211378E-02 then I
calculate KA so KA = 1000 * 1/3.4211378E-02 = 2.923E+04. Do I have to use this EXACT value ? No. The
analysis is only sensitive to orders of magnitude of the value of KA. In the above case, therefore, I would use
KA=3E+4 as the initial KA value. Second, do I have to apply unit loads to EVERY gap GRID in the model (I may
have many, many gaps) ? In general, no. Because I am looking for the order of magnitude of the stiffness for KA, it
is only necessary to select one gap for each region where I judge the stiffness will be significantly different from
other regions. That is to say, if the structure to which the gaps is attached is roughly the same stiffness no matter
which gap GRIDs I select, then one gap will be enough to get a good KA value. Using engineering judgement, I
can see if the structure will have significantly different stiffness in different areas of the model where gaps are
connected. A value of 3E+04 will work just as well as 2E+04 or 4E+04 in the above example. KB - generally I
leave this as default unless I have unusual situations such as a very large value of U0. To illustrate this, take the
example of a gap that has an open value U0=1. I calculate the value of KA to be 2E+16. I leave KB blank, so the
default value becomes KA*1E-14 = 200. Now I apply my loads and eventually during the analysis, the gap closes.
The gap will have moved 1.; gap force = KB*1. = 200. This may be a significant load compared with the applied
load. In this case, I would force KB to 0.1 or smaller to avoid large gap forces due to large axial displacements in
the gap. KT - generally I leave this at default as it is tied to KA via MU1. MU1 and MU2 are self explanatory.
Friction will generally have a significant effect on the cost of the analysis. TMAX, MAR and TRMIN are useful if
the local stiffness of the structure changes significantly as a result of the nonlinear deformations. They allow the
gap stiffness to be adjusted during solution, but only approximate values can be calculated for these. For most
cases, if the structure's stiffness at the gap GRIDs will not change significantly, I would not use adaptive gaps. The
extra iteration of two that maybe needed to get convergence will in general be quicker than reforming the stiffness
to get a not much better value for KA. Only if I expect local stiffness to change significantly would I use TMAX >
0.0. If TMAX is set to 0.0 (the default), the gap-induced stiffness update, gap-induced bisection, and sub-
incremental process are enabled, but the penalty values (KA, KB, KT) remain unchanged. If TMAX is set to a
small positive value, this has all the features of the gap when TMAX=0.0, and also enables the adaptive penalty
capability where the penalty values are adjusted according to changes in the stiffness of the surrounding structure.
A value for TMAX is calculated by examining the structure to which the gaps are attached. If these are shell
elements, then a value of 10% of the thickness of the elements should be used. For other elements, such as beam
elements, then an equivalent thickness is used, such as the depth of the beam in the axial direction of the gap. If the
structure is a massive solid, then the ideal value of TMAX is two or three orders of magnitude less than the elastic
deformation of the solid. This is not always easy to estimate, so a value of 1 * 10E-4 of the characteristic length of
the model can be used. That is, determine the largest dimension of the model and use 10E-4 of that value. If TMAX
277
Structural Engineering Analysis and Design
Second Edition
is too small, the solution will frantically try to adjust the penalty values to update the stiffnesses. If TMAX is too
large; no update will occur and convergence may be elusive. MAR sets the maximum allowable adjustment ratio
for the penalty values and must be in the range 1.0 to 10E6. The upper and lower bounds of the adjusted penalty
values are K/MAR and K*MAR, where K is KA or KT. TRMIN sets the lower bound for the allowable penetration
and is a fraction of TMAX. It must be in the range 0.0 to 1.0. Penalty values are decreased if the penetration is less
than the minimum allowable penetration which is calculated from TRMIN * TMAX. What is allowable ? Back to
good 'ol engineering judgement.
Convergence is often a problem in highly nonlinear problems such as in GAP-like problems. This following
approach is normally a lot faster and more reliable than accepting the default NLPARM entries. The problem with
gap analysis is that the non-linearity is a very severe On/Off situation. This can cause problems where you get
numerical "chatter" when a gap is not open or tightly closed. This chatter makes Nastran think it is not converging
and causes the loads to be bisected. Bisection can be the worst thing to happen, as it reduces the load and can mean
the chatter gets worse. This causes another bisection and before you know it you are applying tiny load increments.
Also, to help whenever a gap opens or closes, we must update the stiffness matrix every iteration. Hence use the
full Newton Raphson. Finally to make the gap closure definite, apply all the load in 1 or 2 steps rather than the
default 10. So in summary use the following settings on NLPARM for a pure gap problem. 1) NINC=1 or 2; 2) Use
Newton Raphson. KMETHOD=ITER, KSTEP=1; 3) Switch off Bisection MAXBIS=0. If convergence is a
problem such as in highly nonlinear problems, then the TSTEP method can always be used with KSTEP=1 such
that stiffness is updated at every time step, hence there is no issue of convergence, just like in explicit nonlinear
transient analysis. However, the time step size must still be small for accuracy.
The CGAP element is not a geometrically nonlinear element, it is a small displacement element, i.e. its
coordinates are not updated as the element deforms. The orientation of the contact plane does not change during
deflection.
The slide line contact elements BLSEG (with BCONP, BFRIC and BWIDTH) are essentially contact curves
that allow significant relative sliding between the contacting parts. Slide lines can also be used to model contact
between surfaces by proper definition of the slide lines. This is done when two or more corresponding slide lines
are positioned on the contacting parts such that they remain relatively coplanar throughout model deformations.
The same stiffness considerations as those for gap elements apply the estimation of the stiffness may be more
complicated than for gap elements. Friction between the sliding interfaces may be incorporated. The friction
coefficient for FEA contact element should not be assumed to have a direct correlation to the friction coefficients
pulled from data sheets. A handy test setup involves a mass on an inclined board supported by a spring.
K U
W sin θ − KU
µ=
W cos θ
W
278
Structural Engineering Analysis and Design
Second Edition
General contact elements are conceptually the simplest but the most computationally intensive. Sets of elements
define the contacting parts. The contact stiffness is usually derived from the material properties of the elements in
the contact pair – typically the Young’s Modulus and in some codes the material damping. The contact and
separation directions are determined from the normals or boundaries of the contacting elements. These contact
entities provide the smoothest and most realistic pressure distribution but the computational price is heavy.
Inelastic, Nonlinear (Limited), Isotropic Material for Beams, Shells and Solids MATS1 (TYPE PLASTIC).
This material model can be used with CBEAM elements (with dedicated plastic hinges at their ends) to perform
plastic collapse analysis of beam frameworks. The elastic-plastic material model requires the definition of the yield
stress. The elastic-plastic material model also requires definition of the yield criteria, i.e. the criteria checked to
initiate yield, namely Tresca, Von Mises, Mohr-Coulomb or Drucker-Prager. The hardening rule must also be
specified, either no hardening (i.e. perfectly plastic), isotropic hardening or kinematic hardening. The isotropic
hardening model does not take the Bauschinger effect into account and hence the compressive yield always equals
the tensile yield. The kinematic hardening model will take into account the Bauschinger effect i.e. the reduction in
the compressive yield once tensile yield has occurred and a stress reversal happens.
Elastic, Nonlinear (General), Isotropic Material for Beams, Shells and Solids MATS1 (TYPE NLELAST).
The nonlinear material model requires the general definition of the stress-strain curve with a TABLES1 entry, i.e.
the multi-linear stress-strain curve. To model a nonlinear elastic cable, a CROD element with the appropriate area
(on a PROD card) can be defined together with a MAT1 and MATS1 card, both sharing the same ID. The MATS1
card references a TABLES1 card which defines the nonlinear stress-strain characteristics of the cable with three
points, for instance (-1.0,0.0), (0.0,0.0) and (1.0,2.05E11) to define a steel tension only cable element.
Hyperelastic Material MATHP. Hyperelastic materials such as rubber, silicone and other elastomer behave
differently than standard engineering materials. Their strain displacement relationship is nonlinear even at small
strains and they are nearly incompressible. The Poisson’s ratio of hyperelastic materials may exceed 0.50, whereas
specifying a Poisson’s ratio of 0.50 or greater will result in failure using any other solution method. If the elements
undergo large net displacements, it is prudent to define hyperelastic elements, otherwise non-equilibrium loading
effects may occur with small strain elements. Because of the corotational formulation where a large deformation is
split into element rigid body deformation and element net deformation, small strain elements can be used to model
large total deformation, but not large net element deformations. A fine mesh will usually ensure that element net
deformations are small. If element net rotations exceed 20 degrees or if the element stretches by more than 10-20%,
then large strain (hyperelastic) elements should be used. Material properties are usually entered as Mooney-Rivlin
constants (consisting of distortional deformation constants and volumetric deformation constants) or as a set of
stress-strain curves (of force versus stretch for simple tension, compression, shear and volumetric compression).
Temperature Dependent Anisotropic Elastic Material for Shell MATT2 and Solid Elements MATT9
279
Structural Engineering Analysis and Design
Second Edition
Buckling (i.e. instability) occurs when the tangent stiffness matrix becomes non-positive definite. To estimate the
buckling load, the following methods can be used.
I. A SOL 106 can be performed until a non-positive definite stiffness matrix is detected. The load steps can be
reduced (by the adaptive bisection method) so the stage at which the tangent stiffness matrix becomes non-
positive definite can be ascertained with greater precision, hence approximating the buckling load.
II. Another method is to use the arc-length method to trace the limit state and the post-buckling path.
III. Another option is to perform linearized buckling analysis at various stages of the equilibrium path based on
the tangent stiffness and differential stiffness at that stage, so-called the linearized buckling analysis based on
a deflected configuration. Note that this is not the true theoretical nonlinear eigenvalue extraction analysis.
However, since the tangent stiffness and the differential stiffness are based on the deflected configuration
from a nonlinear static analysis, we shall refer to this buckling analysis as the nonlinear buckling analysis.
The nonlinear buckling analysis computes a load factor based on the tangent stiffness matrix at not the initial
undeflected configuration, but at a certain deflected position. The eigenvalue extraction method is the usual linear
eigenvalue solution scheme. The nonlinear buckling solution is based on the extrapolation of two consecutive
incremental solutions. The tangent stiffness matrix is proportional to the external loads, which implies that the
critical load may be linearly extrapolated, i.e.
Pcr = Pn + α ∆P
The tangent stiffness matrix is proportional to the displacement increments, hence the critical displacements may be
obtained by extrapolating from the current state, i.e.
Ucr = Un + λ ∆U
Since the tangent stiffness matrix is proportional to the displacement increments i.e. it changes linearly with
displacements, the internal loads are a quadratic function of the displacements.
NASTRAN uses two converged solution points to form the differential stiffness matrix. Hence the buckling
solution tends to be more accurate when these consecutive incremental steps are closer to the buckling point. Since
we do not know where the buckling point is until the analysis is performed, the nonlinear buckling analysis is
performed by dividing the load into a number of subcases and performing the buckling analysis in multiple
subcases based on the tangent stiffness matrix at the current point. Hence, the deck would look exactly the same as
a nonlinear static analysis with
(i) the additional METHOD command in certain subcases referring to the EIGB (with SINV being the
recommended extraction method)
(ii) PARAM, BUCKLE, 2
(iii) KSTEP = 1 in the NLPARM card to force the stiffness matrix to be updated at every solution step
280
Structural Engineering Analysis and Design
Second Edition
$ BULK DATA
PARAM, BUCKLE, 2 $ Buckling Parameter to Activate Nonlinear Buckling Analysis
EIGB ID
To illustrate the nonlinear buckling analysis, if the first subcase has a load totaling 1000 and the second subcase
3000, i.e. an additional 2000. Say, the first subcase refers to a NLPARM entry with 5 increments (i.e. NINC = 5),
hence the incremental load is 200. And say the second subcase refers to a NLPARM entry with 20 increments (i.e.
NINC = 20), hence the incremental load is 2000/20 = 100. Performing a buckling analysis at the end of the first
subcase (by having a METHOD command and PARAM, BUCKLE, 2) solves for α from which the critical
buckling load can be calculated as
Likewise performing a buckling analysis at the end of the second subcase (by having a METHOD command and
PARAM, BUCKLE, 2) solves for another α from which the critical buckling load can be calculated as
Evidently, a better estimate of the buckling load is attained when the equilibrium point from which buckling
analysis is undertaken is close to the buckling point.
Linear modal analysis is performed on the nonlinear static solutions to include effects of prestress and other
nonlinearities. The linear eigenvalue solution analysis is performed based on the tangent stiffness matrix at the end
of the particular subcase or subcases where the METHOD statement appears. The deck would look exactly the
same as a nonlinear static analysis with
(i) the additional METHOD command in certain subcases referring to the bulk data EIGRL or EIGR
entries to select the linear eigenvalue extraction method
(ii) PARAM, NMLOOP, L with the loopid L of any positive value as this is not a restart
$ BULK DATA
PARAM, NMLOOP, L $ Loopid L can be any positive integer as this is not a restart
281
Structural Engineering Analysis and Design
Second Edition
3.4.4.4 Restart From Nonlinear Static Analysis SOL 106 Into Nonlinear Static Analysis SOL 106
Restarts are allowed from converged solutions. A LOOPID is created after each converged load increment in static
analysis. The restart .dat file should contain all the information from the previous analysis with the addition of the
following cards.
$ FMS
RESTART
$ BULK DATA
SUBID is not the subcase ID but the subcase sequence number, which will be the subcase ID if the subcases have
been incremented by 1 from 1. The SUBID value should be incremented by one from the last value printed in .f06.
The LOOPID should be the last value stored for restart as printed in the .f06. On top of the above additional cards,
additional cards which define the restart should be included. The Case Control Section should have the additional
subcases starting with a subcase with the SUBID. The following changes are allowed in the model: -
(i) Additional applied loads, naturally
(ii) Changes to the boundary conditions i.e. SPC may be removed
(iii) Addition of direct input matrices
(iv) Changes to the grid points
(v) Changes to the element, i.e. linear elements may be added or deleted, but should be drastic as to
cause large initial imbalance of loads; also elastic material properties MAT1 may be changed.
3.4.4.5 Restart From Nonlinear Static SOL 106 Into Linear Solution Schemes SOL 107 to SOL 112
It is imperative to realize that linear restarts are incremental values with respect to the preload, not total values.
Hence, the total values are obtained simply by adding the linear restart values to that obtained from the converged
nonlinear static solution. It is advised to have KSTEP = 1 so that the tangent stiffness matrix is computed at every
solution step and to ensure that the linear restarts are based on the very latest tangent stiffness matrix. The deck
should obviously refer to the linear analysis being performed and not SOL 106.
$ FMS
RESTART
$ BULK DATA
PARAM, NMLOOP, L $ Load step id at the end of the subcase from the SOL 106 run
282
Structural Engineering Analysis and Design
Second Edition
3.5.1 Dynamic Relaxation of the Explicit Finite Difference Scheme Solving Newton’s Dynamic
Equilibrium ODE (LS-DYNA)
In dynamic relaxation analysis it is assumed that the loads are acting on the structure suddenly, so the structure is
excited to vibrate around the equilibrium position and eventually come to rest on the equilibrium position. Within
the solution scheme, it is similar to transient analysis except that
(i) artificial viscous damping is applied at each time step
(ii) mass scaling (artificial addition of mass) is employed
Termination occurs when the static solution is reached. If there is no damping applied to the structure, the
oscillation of the structure will go on forever. Therefore, artificial damping is required in the form of reducing the
calculated velocities of all the nodes by a factor at each timestep to allow the vibration to come to rest, and
converge to the static solution.
Loads and prescribed motions (such as displacements, velocities and accelerations) applied during the dynamic
relaxation should always be ramped up from zero in order to reduce the oscillations. Instantaneously applied loads
will induce over-excitation of higher modes, thus lengthening the time taken to converge and possibly introducing
errors. In reality, the static loads are applied in an infinite time interval so as to not induce any dynamic effects.
Hence, the load should be ramped up slowly (in a duration greater than the fundamental period) so as not to induce
considerable dynamic effects.
If there are non-linear materials in the model, load curves which ramp up suddenly could induce large dynamic
amplifications that could cause the non-linear materials to yield, when in reality they should remain elastic in the
application of static loads. If any part of the model is required to yield (i.e. become non-elastic) as part of the
equilibrium position then great care needs to be taken. This is a path-sensitive solution and the deformation should
progress in a monotonic fashion, otherwise the degree of yielding will be incorrect. If all the material models within
the finite element model are linear elastic, then the static solution (by dynamic relaxation) is path independent. As
with all nonlinear static solutions that are based on solving the dynamic equilibrium equations, the solution is only
valid if it is path independent, which in theory is only applicable to models limited to linear material properties. In a
283
Structural Engineering Analysis and Design
Second Edition
true pseudo-static solution, the loading, deflections and the plastic behaviour increase monotonically. But the
dynamic relaxation solution scheme will probably yield a sufficiently accurate solution so long as the dynamic
oscillations are small at every stage of the analysis such that the variations of displacement (and hence elemental
strain) are small so that excess plasticity is not generated. Dynamic relaxation can thus be used in models with
nonlinear materials, but the time of application of the forcing function should be gradual so that the inertial part of
the dynamic equilibrium equation does not incorrectly cause large displacements to occur and incorrectly cause the
nonlinear materials to yield. If this happens, the material yields due to the dynamic effects of the solution scheme
and not due to the underlying static loading, the latter of which is perfectly acceptable. The matter is further
complicated if the material model have unloading curves as well as the plastic regions. To avoid the undesirable
dynamic effects, there should thus be many time steps between each load increment so that there is plenty of
opportunity for the velocities and hence accelerations to be damped down. Recall that artificial damping is applied
at each time step. The time over which the loads are ramped up should aim to gently accelerate the model
deformation and is dependent on the model. The time should not be so short that the model does not have time to
react, effectively making the loading instantaneous. However, the time should not be too long, which would
increase the time taken to converge. This is because the convergence parameter is based on the initial to the current
kinetic energies. If even the initial kinetic energy in the model is too small (as a result of too gently accelerating the
model) then for a certain convergence ratio, the current kinetic energy will have to be much smaller. A duration
slightly greater than the fundamental period is appropriate.
As we are not interested in the dynamic response (since this is a static solution), the mass matrix can be artificially
formed so as to maximize the time step which is limited by the lowest natural period in the model divided by π.
Hence, mass scaling can be employed if damping is not sufficient to produce a converged static solution quickly.
Mass scaling will also reduce the dynamic effects as the acceleration become smaller.
Another method of applying the artificial viscous damping is using DAMPING_GLOBAL which applies global
viscous damping proportional to the mass, i.e. damping force = c.v = 2ζmω.v = 2ζm2π/T.v = 4ζπ/T.m.v. The user
specifies 4ζπ/T where T corresponds to the natural frequency of the most prominent mode excited by the
excitation.
Ideally, critical damping will provide the most efficient dynamic relaxation. However, experience has shown that it
is usually better to under-damp the model. An under-damped model will move just past the equilibrium position
and then rebound to converge on the correct solution. This process is relatively easy to monitor and the equilibrium
position becomes obvious. However, a critically damped model will not rebound, since it moves directly to the
equilibrium position, and thus can often look similar to an over-damped model. Experience shows that damping in
the order of about 5% of critical is about optimum and not much higher.
If the equilibrium position is path-sensitive (i.e. certain materials are required to yield), then ideally a critically
damped model should be used. Alternatively, use plenty of time steps within each load increment so that the
dynamic effects are very much reduced.
284
Structural Engineering Analysis and Design
Second Edition
Ensure that all parts of the model have a velocity less than the terminal velocity. If parts of the model need to
displace a relatively large distance before reaching equilibrium, then the model may reach terminal velocity. This is
when the reduction in velocity due to relaxation damping just cancels out the increase in velocity due to loading
(i.e. no resultant acceleration). This causes dynamic relaxation to take a long time because Vterminal limits the speed
of deformation. To overcome this, ramp up the loading to greater than the real loading (2-3 times gravity) to
deform faster, thus increasing Vterminal, then as equilibrium position reached, ramp load back down, ensuring no
additional modes of deformation occurs that would not have occurred under normal magnitude loading.
Vterminal = F * dt * k / (m*(1-k))
F = desired fraction of critical (normally 1.0)
dt = timestep
m = mass of body
k = damping constant
The effectiveness of the dynamic relaxation solution scheme is investigated by plotting the variation of the
convergence parameter as shown below.
Time
Convergence Parameter C
(a)
(c)
(b)
(d)
The most common problem with dynamic relaxations is that the solution fails to converge or appears to take too
long to converge. Referring to the plot above, we can identify 4 cases
(a) Solutions that never start to converge or take too long to converge. This can be caused by severe
over-damping or by an insufficiently resisted (but loaded) mode of deformation. Apply a lower
damping constant during the dynamic relaxation phase
(b) Solutions that partly converge but never reach the convergence tolerance. This can be caused by
hourglassing or high modes of oscillation such as single element modes. To solve these problems,
try using a different hourglass control or significantly reduce the timestep by say a factor of 5 for
the dynamic relaxation phase
(c) Severe under-damping
(d) The correct solution
285
Structural Engineering Analysis and Design
Second Edition
To specify termination time of zero for the explicit dynamic analysis to have no limit
*CONTROL_TERMINATION
Termination
Termination
Termination Change in
Termination Termination Min Time
Energy Ratio Total Mass
Time Limit Cycle Limit Step Size
Limit From Mass
Limit
Scaling
To set structural timestep size control and invoke mass scaling if necessary
*CONTROL_TIMESTEP
Min Time
Maximum
Initial Time Factor For Shell Min Step For Erosion Limit Mass
Basis Time Step
Step Time Step Time Step Mass Flag Scaling
Curve
Scaling
To specify user-defined boundary conditions for the dynamic relaxation (stress initialization) phase,
*LOAD_BEAM_<ELEMENT, SET>
*SET_BEAM
*DEFINE_CURVE with SIDR = 1 for stress initialization phase
*LOAD_RIGID_BODY
*DEFINE_CURVE with SIDR = 1 for stress initialization phase
To apply automatically computed gravitational loads, noting that by d-Alembert’s principle, applying accelerations
in a certain direction results in inertial loads acting in the opposite direction,
286
Structural Engineering Analysis and Design
Second Edition
To prescribe displacements boundary conditions on a node, on a set of nodes, on a rigid body in the global axes
system or on a rigid body in the rigid body local axes system,
*SET_NODE_LIST
NSID DA1 DA2 DA3 DA4
NID NID NID NID NID NID NID NID
NID NID NID NID … … … …
*DEFINE_CURVE
LCID SIDR SFA SFO OFFA OFFO DATTYP
Abscissa Values Ordinate Values
Abscissa Values Ordinate Values
… …
SIDR defines whether the curve is valid for the transient phase, stress initialization phase or both. Here obviously,
SIDR = 1. The curve should define a constant value of displacement valid throughout the analysis. If the dynamic
relaxation phase is to be followed by a transient seismic analysis, it is necessary to define zero velocity prescribed
boundary conditions for the dynamic relaxation phase on the nodes at which the transient accelerations are applied
so that the structure is supported during dynamic relaxation.
287
Structural Engineering Analysis and Design
Second Edition
The natural frequencies of a structure are the frequencies at which the structure naturally tends to vibrate if
subjected to a disturbance. Hence modal analysis warns if the dominant forcing frequencies are close to the lowest
natural frequencies of the structure, indicating resonance behavior of the lowest modes of the structure.
288
Structural Engineering Analysis and Design
Second Edition
Note that the value of the generalized (or modal) mass and generalized (or modal) stiffness is arbitrary since it is
dependent upon how the corresponding eigenvector has been normalized. However, for a particular mode,
K i = ω ni Mi
2
[ ]
modal stiffness matrix, [K] = [Φ ] [K ][Φ ] = ω 2 [M]
T
A normal modal analysis indicates the best location for the accelerometers in dynamic testing. Design changes can
be evaluated with a normal modal analysis, in that if a particular modification were to cause a change in the
frequencies and mode shapes, then the response in also likely to change. This is done with the knowledge that the
natural frequencies are a function of the structural properties and boundary conditions but the mode shapes are a
function of only the boundary conditions. Modal strain energy is a useful quantity in identifying candidate elements
for design changes to eliminate problematic low frequencies. Elements with large values of strain energy in a mode
indicate the location of large elastic deformation (energy). Stiffening these elements will increase the natural
frequency more than stiffening other elements.
It must be understood that the scaling (or normalization) of the normal modes are arbitrary, hence their do not
indicate the response. This means that a different scaling method will yield a different modal (or generalized) mass,
modal stiffness (but same modal frequency) and different modal response. However the response due to the
particular mode in the physical coordinates will be unique irrespective of the method of scaling.
Even the comparison of magnitude of response between the different mode shapes or different modal masses in
itself to determine relative importance of each mode is incomplete (as the amplitude and location of the excitation
i.e. the modal mass is also important), although the response on different parts of the structure due to a particular
289
Structural Engineering Analysis and Design
Second Edition
mode can be indicative of the more stressed parts due to that particular mode. But it remains that the mode shapes
only represent the shapes of vibration that the structure vibrates in, and not the magnitude, which is dependent upon
the amplitude and location of loading and the relation between the forcing frequencies and the natural frequencies.
The orthogonality of normal modes signifies that no mode can be obtained from a linear combination of other
modes. The representation of a normal mode by using a modal mass and modal stiffness is useful for formulating
equivalent dynamic models and in component mode synthesis.
Rayleigh’s quotient as presented above is analogous to the equation of natural circular frequency ωn2 = K/M for a
SDOF system, here instead the natural circular frequency of the mode is the division of the modal stiffness to the
modal mass, square rooted.
Identical eigenvalues (repeated roots) occur in structures that have a plane of symmetry or that have multiple
identical pieces (such as appendages). The eigenvectors (mode shapes) for these repeated roots are not unique
because many sets of eigenvectors can be found that are normal to each other. Consequently, small changes in
model or different computers can make large changes in the eigenvectors for the repeated roots. Rigid body modes
are a special case of repeated roots.
The modal dynamic analysis is based on only the instantaneous stiffness matrix at the initial undeflected stage. The
modal dynamic P-∆ (KGA From KTA) analysis is based on the instantaneous and geometric (or differential) stiffness
matrices at the initial undeflected stage i.e. taking in to account the effect of the initial prestress and initial non-
work conjugate external load on the stiffness of the structure in the computation of the structural frequencies and
mode shapes.
Modal Dynamic [[ K AE ] − ω ni [ M ]]{φ}i = {0}
2
It is worth noting that only the stiffness and the mass distribution of the structure determine its modes. The static
equilibrium displacement does not come into the equations. However, if geometrical non-linearity is taken into
290
Structural Engineering Analysis and Design
Second Edition
account, the tangent stiffness of the structure will change from the initial undeflected configuration to its static
equilibrium configuration, in which case the real modal analysis will be affected.
It is important to check for rigid body motion. The rigid body modes involve translation or rotation of the model,
without any deformation i.e. stress free conditions indicating modelling errors is an inadequate constraint set or the
analysis of unconstrained structures such as airplanes. Rigid body modes have a zero frequency. In practice, due to
roundoff errors the frequency of these modes is a small number. The modes do not occur in such a pure form and
are often coupled together in some way but there should always be 6 rigid-body modes and their natural
frequencies should be close to 0.0 Hz (~1.0E-4) as their modal mass is greater than zero whilst their generalized
stiffness is zero. The user should check that this number is several orders of magnitude lower than the lowest
natural frequency. A zero frequency mode which does not also have zero strain energies is not a rigid body mode,
instead could be a mechanism and as such should be investigated. The number of rigid body modes depends on the
way the structure is supported. For example, an unsupported vehicle model would produce 6 rigid body modes (3
translations, 3 rotations), a model of half the vehicle with symmetry boundary conditions should produce 3 rigid
body modes. A cantilever beam should have no rigid body modes. A common modelling error resulting in a
mechanism is when a bar is cantilevered from a solid element; the bar has rotational stiffness and the solid has no
rotational stiffness, resulting in a pinned connection when the two are joined. The user should check the natural
frequency is of the right order of magnitude. For example, the lowest bending and torsion modes in a standard
saloon vehicle are 20-30 Hz, while the lowest modes in multi-story buildings are 2-5 Hz. The natural modes should
be animated, and the overall behavior should be assessed. It is important to distinguish between fundamental modes
involving the whole structure, and local modes involving limited zones. The local modes can often be due to
modelling problems (for example flapping panels in vehicle structures), and a judgment should be made whether
they are unrealistic.
Modal analysis is extremely useful to check for unrestrained degrees-of-freedom. Note that in this case the analysis
will often run but spurious mode shapes will often result involving a component that has not been properly attached
to the main structure. It is often true that merely by post-processing the results from a modal analysis and studying
the mode shapes most of the problems concerning the connectivity of the FE model become apparent.
Alternatively, the edges of the finite element mesh can also be highlighted with a preprocessor and this could
provide useful insight.
A matrix whose eigenvalues are all greater than zero is said to be positive definite. If the eigenvalues are zero and
positive, the matrix is positive semi-definite. An unconstrained structure would have a positive semi-definite
matrix. If all the rigid-body modes are constrained, the stiffness matrix will then be positive definite.
An extremely reliable mathematical tool that determines the number of eigenvalues that exists within a certain
frequency range is the Sturm Sequence Check. This is printed under UIM 5010 and should be inspected always.
Modal testing and analysis correlation is done using orthogonality checking and Modal Assurance Criteria (MAC)
using the postmaca.v2001 DMAP alter. It allows for the quick assessment of modal vectors obtained from test data.
The orthogonality check indicates if a set of vectors is orthogonal to the mass matrix, and hence shows if the large
mass terms have reasonable agreement between test and analysis. The MAC indicates a set of vectors is linearly
independent, and hence indicates how well the modal vectors from the test and analysis agree with each other. As a
general rule for the two checks, the diagonal terms should be greater than 0.9 (with lower order modes very close to
unity) and off-diagonal terms should be less than 0.1. Note that both the analysis DOFs and test accelerometers
should be in the same location and orientation.
291
Structural Engineering Analysis and Design
Second Edition
SOL 103
$ CASE CONTROL SECTION
$ Sets defining grid ids or element ids
SET < Number > = 1 THRU 100, 211, 343, < etc >
$ Grid output of displacement for each mode i.e. eigenvector
$ SORT1 lists the results by eigenvalue whilst SORT2 lists the results by grid id
DISPLACEMENT(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>) = ALL/<Grid Set ID>
$ Grid output of real eigenvector for the a-set
SVECTOR(<PRINT,PUNCH>) = ALL/<Grid Set ID>
$ Grid output of SPC forces
SPCFORCES(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>) = ALL/<Grid Set ID>
$ Element output of force, stress and strain
ELFORCE(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
ELSTRESS(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
STRAIN(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>) = ALL/<Element Set ID>
$ Analysis Cards
SPC = < ID of SPC Cards Defined in Bulk Data >
METHOD = < ID IN EIGRL or EIGR >
$ XY plot output
OUTPUT(XYPLOT)
XYPUNCH <DISP/VELO/ACCE> RESPONSE <subcase>/<Grid ID>(<T1/T2/T3>)
XYPUNCH <ELFORCE/ELSTRESS/STRAIN> RESPONSE <subcase>/<Element ID>(<Code Number>)
The recommended method of eigenvalue extraction is the Lanczos method, which combines the best
characteristics of both the transformation (Givens, Householder, Modified Givens, Modified Householder) methods
and iterative tracking (Inverse Power, Strum Modified Inverse Power) methods. The Lanczos method requires that
the mass matrix be positive semi-definite and the stiffness be symmetric but does not miss modes. The Lanczos
method is defined with an EIGRL entry which allows either MASS (default) or MAX eigenvector normalization.
$ BULK DATA
Lower Upper Eigenvector
Number of
EIGRL ID Frequency Frequency MSGLVL MAXSET SHFSCL Normalization
Eigenvalues
(Hz) (Hz) Method
A backup method that does not miss roots and is useful for finding out the first few modes for very large models
that do not fit in memory is the Sturm Modified Inverse Power Method (SINV), which allows either MASS
(default), MAX or POINT eigenvector normalization.
$ BULK DATA
Lower Upper
Number of
EIGR ID SINV Frequency Frequency
Eigenvalues
(Hz) (Hz)
Eigenvector Grid ID
DOF for
Normalization for
POINT
Method POINT
292
Structural Engineering Analysis and Design
Second Edition
It is often necessary to incorporate the reduction in bending stiffness of gravity load resisting columns for the
analysis of lateral loads. The following procedure is undertaken.
Phase 1
Perform static analysis (with loads that cause the greatest negative or positive geometric stiffness) based on [KEA]
but include the segyroa.v2001 alter prior to the Case Control Section in order to compute the geometric stiffness
matrix [KGA]1 (and output into a .pch file) based on the generated element loads from the [KEA] static analysis.
Phase 2
A SOL 103 is undertaken based on [KEA] + [KGA]1 with the k2gg=ktjj statement in the Case Control Section and the
outputted .pch file which contains the ktjj matrix in the Bulk Data. The responses at this stage represent the P-∆
(KGA From KEA) real modal response.
The method is valid when only the prestress is judged to affect the geometric stiffness such as in the
compressive preload of building columns due to gravitational loads and the prestressing of extremely taut cables
that sag very little under gravity but not in systems such as suspension bridges. Where lateral loads are large
enough to affect the geometry of the system with prestress, then a repetitive P-∆, SOL 106, implicit dynamic
relaxation SOL 129 or explicit dynamic relaxation must be employed. But in single P-∆ analysis, because cables do
not have much elastic bending stiffness, the initial static preload subcase should only include the prestress and not
gravity as including gravity is the same as solving two linear static problems of stiffness KEA with preload and
gravity as the applied loads respectively. Clearly, in the gravity case, it is nonsensical as the cables do in reality
have differential stiffness (from the prestress) to resist the gravitational force. Prestress in one direction (i.e. along
the axis of cable) will cause a differential stiffness in the orthogonal direction. Gravity acts in the orthogonal
direction and hence cannot be accounted for in the calculation of the prestress in this single P-∆ analysis. To
quantitatively decide if gravity need not be considered in contributing to the differential stiffness of the cables, a
static P-∆ analysis should be carried out, the first subcase being a SOL 101 with only the prestress as applied loads
and the second subcase a P-∆ SOL 101 (i.e. utilizing the induced prestress from the first subcase to form a
geometric stiffness matrix) with both the gravity and prestress included as applied loads. If the difference in the
cable element forces between subcases 1 and 2 is negligible, then gravity has little influence in affecting the
geometric stiffness. If there is a major difference in the cable element force, then clearly, gravity will affect the
geometric stiffness and as such, a repetitive P-∆ , SOL 106, implicit dynamic relaxation or explicit dynamic
relaxation must be used to converge to the true KT. Likewise, in the single P-∆ analysis of multi-storey buildings,
gravity (and only gravity) acts in axis of columns to generate prestress, and the differential stiffness is computed for
the orthogonal direction reducing resistance to lateral wind forces, applied in the second subcase with gravity too.
293
Structural Engineering Analysis and Design
Second Edition
The STATSUB(PRELOAD) computes the differential stiffness due to the prestress and also the follower force. The
follower force is calculated and incorporated by the use of PARAM, FOLLOWK, YES. We know how the prestress
affects the differential stiffness, namely a tensile prestress causing an increase in stiffness. The effect of the follower
force on the stiffness is different. For example, for a cylinder under external pressure critical buckling load may be over-
estimated (even though the mode shapes are similar) in a SOL 105 and the natural frequencies in vibration may be under-
estimated (even though the mode shapes are similar) in a SOL 103 in the absence of follower stiffness. To the contrary,
this observations are reversed in case of centrifugal loads. Centrifugal forces as a constant (static) load are applied by a
Bulk Data RFORCE to any elements that have masses. The follower stiffness due to centrifugal load has the effect of
lowering stiffness (although the centrifugal load tensioning effect increases stiffness), consequently lowering natural
frequencies (even though the mode shapes are similar) in a SOL 103 and lowering the buckling loads (even though the
mode shapes are similar) in a SOL 105. This effect increases as the RPM increases, and it becomes significant when the
RPM is over 1000. For moderately geometric nonlinear analysis, exclusion of follower stiffness affects the rate of
convergence, but the converged solution is correct. For severely geometric nonlinear analysis, it may not be
possible to obtain a converged solution without including follower stiffness. As the geometric nonlinearity
intensifies, so is the effect of follower stiffness. Therefore, inclusion of follower stiffness greatly enhances the
convergence if the deformation involves severe geometric nonlinearity.
294
Structural Engineering Analysis and Design
Second Edition
4.1.2.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Real Modal Analysis
It is often necessary to include the differential stiffness, especially if there are prestressed cables in the model and
the mode shapes are sought.
Modal Dynamic P - ∆ [[ K TA ] − ω ni [M ]]{φ}i = {0}
2
To obtain KTA, to be theoretically exact, a GNL SOL 106 (or implicit dynamic relaxation SOL 129 or explicit
dynamic relaxation LS-DYNA) with prestress (as temperature loads say) and gravity must be undertaken.
Alternatively, an approximation to KTA can be obtained by repetitive P-∆ static analyses with the prestress (as
temperature loads say) and gravity applied. The procedure to obtain this approximate KTA will be presented. Note
that the approximate KTA will be the summation of the elastic stiffness KE at the undeflected (by the prestress and
gravity) state but KG at the deflected (by the prestress and gravity) state. Hence if KE changes considerably during
the application of the prestress, a full SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic
relaxation LS-DYNA), which converges to the KE and KG at the deflected (by the prestress and gravity) state
should be employed. Hence for the modelling of a suspension bridge where there is a great change in geometry
(known in the bridge industry as form-finding, so-called because it is necessary to find the form or shape of the
catenary suspension cables), it may be prudent to employ SOL 106 (or implicit dynamic relaxation SOL 129 or
explicit dynamic relaxation LS-DYNA), but for a high tension low sag cable on say a tower with prestressed
cables, the repetitive P-∆ static analysis may be adequate. The repetitive P-∆ analysis basically involves a number
of iterations of linear static analyses to obtain an approximate KTA. Note again that A refers to the initial
undeflected (by the collapsing load) state, but deflected by the prestress and gravity. To perform the repetitive P-∆
analysis, a static analysis is performed based on KEA with temperature loads and gravity to generate forces in the
structural elements, which in turn provides input for the computation of KGiAKTm where m is the iterations.
Repetitive static analysis is performed with the prestress and gravity updating the stiffness matrix KEA + KGiAKTm-1 +
KGiAKTm until convergence of displacements is obtained. The tangent stiffness at this stage is the approximate
converged tangent stiffness matrix KTA = KEA + KGiAKT. The converged displacements represent the approximate P-
∆ (KGA From Approximate KTA) static response to the initial prestress loads. The converged geometric stiffness at
this stage would be that based upon the approximate tangent stiffness matrix KTA, i.e. KGiAKT.
Phase 1
Perform static analysis (with prestress and gravity) based on [KEA] but include the segyroa.v2001 alter prior to the
Case Control Section in order to compute the geometric stiffness matrix [KGA]1 (and output into a .pch file) based
on the generated element loads from the [KEA] static analysis.
Phase 2
Perform static analysis (with prestress and gravity) based on [KEA] + [KGA]1 by including the k2gg = ktjj statement
in the Case Control Section, the outputted .pch file which contains the ktjj matrix in the Bulk Data and the
segyroa.v2001 alter prior to the Case Control Section to compute the [KGA]2 (and output into the .pch file
overwriting previous data) based on the generated element loads from the [KEA] + [KGA]1 static analysis.
Phase 3
Repeatedly perform the Phase 2 static analysis (with prestress and gravity) based on [KEA] + [KGA]i for i = 2 to n
where n represents the number of iterations required for the change in deflections between analyses to become
negligible. This would signify that the change in the [KGA] matrix become negligible and the correct [KGA] is
attained. The deflections and the other responses at this stage represent the P-∆ (KGA From Approximate KTA) static
response to the prestress and gravity. The stiffness of the structure is KTA.
Phase 4
A SOL 103 is undertaken with the k2gg=ktjj statement in the Case Control Section and the outputted .pch file
which contains the latest ktjj matrix in the Bulk Data. The responses at this stage represent the P-∆ (KGA From
Approximate KTA) real modal response.
295
Structural Engineering Analysis and Design
Second Edition
Equation of motion
m&u&( t ) + ku ( t ) = 0
Assume general solution
u(t) = Ge λt where G = G R + iG I and λ = α + iωd
Hence
mGλ2 e λt + kGe λt = 0
(mλ 2
)
+ k Ge λ t = 0
for LHS to be zero for all t
(mλ )
+ k = 0 as
2
Ge λt > 0
the roots of this quadratic characteristic equation are
k k
λ1 = +i λ 2 = −i
m m
k
∴α = 0 and ωd = ω n = rad / s
m
note that the relationships,
ωn 2π
natural frequency, f n = Hertz and the period, Tn = s
2π ωn
hence, the complementary function,
u(t) = G1e λ1t + G 2 e λ 2 t
u(t) = (G1R + iG1I )(cos ωn t + i sin ωn t ) + (G 2 R + iG 2I )(cos ωn t − i sin ωn t )
u ( t ) = [((G1R + G 2 R )cos ωn t − (G1I − G 2 I )sin ωn t ) + i((G1I + G 2 I ) cos ωn t + (G1R − G 2 R )sin ωn t )]
the free vibration response must be real for all t, hence
G1I = −G 2 I and G1R = G 2 R
sin ce there are two less independent constants, let
G1I = −G 2 I = G I and G1R = G 2 R = G R
we notice that G1 and G 2 are a complex conjugate pair
G1 = G R + iG I and G 2 = G R − iG I
hence the general solution
u ( t ) = 2G R cos ωn t − 2G I sin ωn t
Substituti ng the integration constants with the initial conditions,
u&(0)
u ( t ) = u (0) cos ωn t + sin ωn t
ωn
Hence the maximum dynamic displacement
2
u&(0)
u max = + u (0) 2
ω
n
296
Structural Engineering Analysis and Design
Second Edition
A free vibrational dynamic response analysis involves initial displacement and velocity conditions but no forcing
vector. It is important to realize that the vibration occurs about the static equilibrium position. Hence we always
define u(t) to begin from the static equilibrium position. However, the dynamic analysis need not begin at this
point. For example, in the instance of a mass m on a string of tensile stiffness k dropped from a level where the
string is not taut, the dynamic analysis begins (t = 0.0) when the string just becomes taut and hence u(t = 0.0) =
−mg/k and du/dt (t = 0.0) will be the velocity of the mass when it reaches that particular taut string elevation.
Likewise, when a mass m is dropped h meters onto the tip of a cantilever, u(t = 0.0) = −mgL3/(3EI) and du/dt (t =
0.0) = (2gh)1/2. The static displacement was found simply from the force mg divided by the stiffness 3EI/L3.
From the knowledge of the initial conditions, we can determine the maximum dynamic displacement by solving the
free vibrational dynamic equation of motion. In summary, to determine the maximum dynamic displacement,
(i) first identify the u(t) = 0 reference as the position of static equilibrium;
(ii) second, identify the t = 0 level as when dynamic analysis begins, i.e. when a dynamic equation of
motion can be written down
(iii) evaluate the initial conditions u(t = 0.0) and du/dt (t = 0.0)
(iv) evaluate the natural circular frequency of the system
(iv) the maximum dynamic displacement can be ascertained from the formula
To estimate the undamped natural circular frequency, we employ ωn=(k/m)0.5. The single mass m is readily
obtainable for SDOF systems. The elastic stiffness in the direction of the dynamic freedom k is obtained from
either of the following methods.
4.1.3.1.1 Natural Frequency Using the Unit Load Method (Virtual Work)
If αs > 0 and given BMD for non-unity external action, say for a load of 700. 1’∆ = f1’δ expresses the unit load
method. δ700 is given, and to find f1, simply randomly release any indeterminacy to draw the virtual internal action
297
Structural Engineering Analysis and Design
Second Edition
diagram f1 removing the real external load 700 and applying the virtual external unit load. Then, ∆700 = f1’δ700 and
finally, k = 700 / ∆700. To estimate the undamped natural circular frequency, we employ ωn=(k/m)0.5.
It is worth noting that the equivalent stiffness of springs in parallel is the addition of the individual stiffnesses and
the equivalent stiffness of springs in series is the inverse of the addition of the inverse of the individual stiffnesses.
Stiffness in series occurs when the load path has to travel both stiffnesses sequentially. Stiffness in parallel occurs
when the load path is given the option to be more attracted to the stiffer element. In series, there is no question of
whether the load will travel the path or not, for it must, all loads must travel some path, and in series, there is no
other option. A mass on the bottom of a vertically hanging spring at the free end of a bending beam cantilever can
be idealized as a SDOF system with equivalent springs in series.
1 1 L3
= +
k e k spring 3EI
Single story shear frames can be idealized as SDOF systems with the stiffness from the bending beam columns
equivalent to a system of springs in parallel because of the axial rigidity of the horizontal member of the shear
frame. By definition, there is full moment rigidity between the top of the columns and the rigid horizontal member.
Standard fixed ended bending beam lateral stiffness used for shear frames with fixed bases is k = number of
columns x 12EI/L3. Standard cantilever bending beam lateral stiffness used for shear frames with pinned bases, k =
number of columns x 3EI/L3. An example of a shear frame includes water tank towers.
4.1.3.1.3 Natural Frequency Using the Static Deflection (used when EI not readily available)
g k g k g 1
fn = = = = 15.76 u st in mm
2π mg 2π W 2π u st u st
Which now relates the natural frequency to the static deflection ust of a single lumped mass dynamic DOF. This
formula is exact for single lumped mass systems.
298
Structural Engineering Analysis and Design
Second Edition
Either the stiffness or flexibility formulation to the equation of motion can be employed. The solution can be
performed by either the exact classical determinant method, the approximate Stodola iterative method, the
approximate Rayleigh Quotient method or the approximate Rayleigh-Ritz method.
Shear frames analysis assumes that the mass of the structure is concentrated at the floors, that the axial force in the
columns are too small to consider axial deformation, and that the floor girders are infinitely stiff compared to the
columns such that there is no rotation at the connection between the floor and the columns. The third assumption
means that the stiffness (or flexibility) of the structure is only provided by the columns.
The free, undamped vibration of the shear frame is described by the equation of motion
The [K] matrix can be the instantaneous stiffness matrix [KEA] or the tangent stiffness matrix [KTA], the A denoting
at its initial undeflected configuration. In the instantaneous stiffness matrix, the stiffness coefficient kij is the force
required at the ith coordinate to maintain a unit displacement at the jth coordinate while all other coordinates remain
stationary. Formulate [KEA] column-by-column (each column corresponding to a unit displacement at the jth
coordinate) noting the forces required in all the i coordinates. Ensure [KEA] is symmetric.
M1
u1
h1 EI1
M2 u2 M1 0 0
[M ] = 0 M2 0
h2 EI2 0
M3 0 M 3
u3
h3 EI3
A unit displacement of u1 would induce a force of 2x12EI1/h13 at DOF 1 and a force of -2x12EI1/h13 at DOF 2 for
all other DOFs to remain stationary. A unit displacement of u2 would induce a force of 2x12EI1/h13+2x12EI2/h23 at
DOF 2, a force of -2x12EI1/h13 at DOF 1 and a force of -2x12EI2/h23 at DOF 3 for all other DOFs to remain
stationary. A unit displacement of u3 would induce a force of 2x12EI3/h33 at DOF 3 and a force of -2x12EI2/h23 at
DOF 2 for all other DOFs to remain stationary. Hence, the instantaneous stiffness matrix,
24EI1 24EI1
3
− 0
h1 h 13
24EI1 24EI1 24EI 2 24EI 2
[K AE ] = − + −
h 13 h 13 h 32 h23
0 24EI 2 24EI 2 24EI 3
− +
h 32 h 32 h 33
Note that the stiffness contributions from the columns at the same level are added because the stiffnesses are in
parallel to each other. Also, had a base been pinned, then the lateral stiffness contribution would just be 3EI/h3
instead of 12EI/h3.
299
Structural Engineering Analysis and Design
Second Edition
It may also be prudent to include the reduction in stiffness from the P-∆ effect, i.e. a linearized geometric stiffness
matrix. A unit displacement of u1 would induce a secondary moment of M1g in the topmost columns (note all the
topmost columns and not just one of them) for all other DOFs to remain stationary. This can be represented by a
force M1g/h1 at DOF 1 and a force -M1g/h1 at DOF 2. Likewise, a unit displacement of u2 would induce a secondary
moment of M1g in the topmost columns and a secondary moment of (M1+M2)g in the intermediate columns. This
can be represented by a force -M1g/h1 at DOF 1, a force M1g/h1+(M1+M2)g/h2 at DOF 2 and a force -(M1+M2)g/h2
at DOF 3. Finally, a unit displacement of u3 would induce a secondary moment of (M1+M2)g in the intermediate
columns and a secondary moment of (M1+M2+M3)g in the lowermost columns. This can be represented by a force -
(M1+M2)g/h2 at DOF 2 and a force (M1+M2)g/h2+(M1+M2+M3)g/h3 at DOF 3. Hence, the geometric stiffness
matrix,
M 1g M 1g
− 0
h1 h1
M1g (M1 + M 2 )g (M1 + M 2 )g
] = − 1
Mg
A
[K Gn + −
h1 h1 h2 h2
0 −
(M1 + M 2 )g (M1 + M 2 )g + (M1 + M 2 + M 3 )g
h2 h2 h3
It is important to point out that the instantaneous stiffness matrix represents the forces required to produce and
maintain the unit displacements in the pertinent DOFs. The geometric stiffness matrix on the other hand represents
the forces which are generated due to the unit displacements in the pertinent DOFs. Hence the equation of motion is
The total solution of the free vibrational problem is given by the linear superposition of the individual modes of
vibration.
A1 sin(ω1 t − α1 )
Μ
{u} = [{φ}1... {φ}i ... {φ}n ] A i sin(ωi t − α i )
Μ
A n sin(ω n t − α n )
The constants Ai and αi are the constants of integration to be determined from the initial displacement and velocity
conditions.
Note that the flexibility matrix [F] of the frame can be obtained from a linear static analysis program by
successively applying unit loads at each storey and on each occasion measuring the total displacement at all the
storey levels. The matrix [F] is then inversed to obtain the stiffness matrix [K].
300
Structural Engineering Analysis and Design
Second Edition
Stiffness formulation,
[[K ] − ω 2ni [ M ]]{φ}i = {0}
1
Multiply by [F] or by − [ F],
ω 2ni
1
[[F][K ] − ω 2ni [ F][M ]]{φ}i = {0} or [[F][ M ] − [ F][K ]]{φ}i = {0}
ω 2ni
1
[[I] − ω 2ni [ F][M ]]{φ}i = {0} or [[F][ M ] − [ I]]{φ}i = {0} ( Flexibility formulations)
ω 2ni
These are eigenvalue problems.
Flexibility coefficient fij is the displacement produced at the ith coordinate due to a unit force applied at the jth
coordinate only. Formulate [F] column-by-column (each column corresponding to a unit force at the jth coordinate)
noting the displacements induced in all the i coordinates. In formulating these columns in [F], 3 cases are
identified: -
(a) Unit force above the displacement of the DOF sought Displacement of the DOF is the flexibility of
all that including and below the displacement DOF
(b) Unit force inline with the displacement of the DOF sought Displacement of the DOF is the
flexibility of all that including and below the displacement DOF
(c) Unit force below the displacement of the DOF sought Displacement of the DOF is the flexibility of
all that including and below the unit force
In short, the displacement at the particular DOF sought after is the cumulative flexibility (inverse of stiffness) of all
the shear frame levels including and below the displacement DOF or the unit force, whichever is lower. Ensure F is
symmetric.
M
u1 u1 1 9
3 3 =
3m EI 12EI/3 12EI/3 12EI / 3 + 12EI / 3
3 3
8EI
M u2 u2
1 9
12EI/33 12EI/33 =
3m EI 12EI / 3 + 12EI / 3
3 3
8EI
M u3 u3
1 9
3m 2EI 12(2EI)/33 12(2EI)/33 =
12(2EI) / 3 3 + 12(2EI) / 3 3 16EI
301
Structural Engineering Analysis and Design
Second Edition
This equation has one additional unknown and hence does not have a unique solution. For a 3 DOF dynamic
system, there will be 3 φ terms and the ωni2 term, i.e. one additional unknown. For there to be a solution that is
nontrivial (i.e. that not all φ terms zero), the determinant must be zero. The eigenvalue problem is solved for ωni2 by
employing the classical determinant computation for a nontrivial solution,
This leads to a polynomial in ωni2 the order of which corresponds to the number of dynamic degrees of freedom.
For instance, for a 3 DOF dynamic system, the order of polynomial in ωni2 will be cubic, i.e. up to the ωni6 term.
They can be solved for the 3 natural circular frequencies ω1, ω2 and ω3. Then, in order to obtain the an eigenvector
corresponding to each eigenvalue, we replace each eigenvalue in turn into
Now that we have found ωni2 based on a zero determinant, the number of independent equations becomes one less,
i.e. for the 3 DOF dynamic system, 2 independent equations. Hence for each ωni2 we can solve for the relative
values of φ. This is where we have to decide on how to normalize the eigenvectors, i.e. whether MAX, MASS or
POINT normalization.
Note that the first mode or fundamental mode refers to the mode corresponding to the lowest frequency. The other
modes are referred to as higher harmonics.
302
Structural Engineering Analysis and Design
Second Edition
With this method, only the fundamental natural circular frequency and the fundamental mode can be obtained fairly
accurately. Higher natural circular frequencies and modes can also be obtained inefficiently with the use of the
sweeping matrix to purify the eigenvalue problem off the earlier modes of vibration. However, this method is
subjected to cumulative errors and so the errors increase in estimating the frequencies and mode shapes of higher
modes of vibration.
Stiffness formulation,
[K − ω 2 M ]φ = 0
1
Multiply by F or by − F,
ω2
1
[FK − ω 2 FM ]φ = 0 or [FM − FK ]φ = 0
ω2
1
[I − ω 2 FM ]φ = 0 or [FM − I]φ = 0 ( Flexibility formulations)
ω2
Employ the latter flexibility formulation,
1
[FM − I]φ = 0
ω2
1
FM φ = φ
ω2
Stodola ' s method,
1
FM φ = φ is akin to A x = λ x
ω2
(0 )
Guess x
For i = 0 to n
(i )
Compute A x
( i +1)
to obtain λ(i +1) and x
(i )
Normalise first coordinate of A x
Next i
1 ( n +1)
The final λ(n +1) is the and x the fundamental mode φ1 .
ω12
303
Structural Engineering Analysis and Design
Second Edition
M
u1
1 9
12EI/33 =
3m EI 12EI/33 3
12EI / 3 x 2 8EI
M u2
1 9
3 3 3
=
3m EI 12EI/3 12EI/3 12EI / 3 x 2 8EI
1.5M u3
1 9
=
3m 2EI 12(2EI)/33 12(2EI)/33 3
12(2EI) / 3 x 2 16EI
2M
u4
1 8
4m 4EI 3(4EI)/43 3(4EI)/43 3
=
3(4EI) / 4 x 2 3EI
Displacement contribution of
Transverse stiffness floor level due to unit load at
of columns the particular level
The figure above shows a 4-degree of freedom shear frame. The flexibility formulation is described by
1
[FM − 2 I]φ = 0
ω
1
FM φ = φ
ω2
Stodola ' s method,
1
FMφ = 2 φ is akin to A x = λ x
ω
(0)
Guess x
For i = 0 to n
(i )
Compute A x
( i +1)
to obtain λ(i +1) and x
(i )
Normalise first coordinate of A x
Next i
1 ( n +1)
The final λ(n +1) is the and x the fundamental mode φ1 .
ω12
9 9 9 8 9 9 8 9 8 8
8EI + 8EI + 16EI + 3EI + +
8EI 16EI 3EI
+
16EI 3EI 3EI
9 9 8 9 9 8 9 8 8 263 209 155 128
+ + + + + 209 155 128
F = 8EI 16EI 3EI 8EI 16EI 3EI 16EI 3EI 3EI = 1 209
9 8 9 8 9 8 8 48EI 155 155 155 128
+ + +
16EI 3EI 16EI 3EI 16EI 3EI 3EI
8 8 8 8 128 128 128 128
3EI 3EI 3EI 3EI
M 0 0 0
0 M 0 0
M= 3
0 0 M 0
2
0 0 0 2M
304
Structural Engineering Analysis and Design
Second Edition
Hence,
263 209 155 128 M 0 0 0
0
1 209 209 155 128 0 M 0
A = FM = 3
48EI 155 155 155 128 0 0 M 0
2
128 128 128 128 0 0 0 2M
263 209 232.5 256
209 232.5 256
M 209
=
48EI 155 155 232.5 256
128 128 192 256
For higher modes the sweeping matrix can be employed. This method is computationally intensive and the rate of
convergence depends very much on whether the initial guessed mode shape approximates the actual mode shape
sought after. Also the errors are cumulative as the sweeping matrix is a function of the previous mode shape.
Hence, higher modes are less accurate. As will be demonstrated hereafter, the first and second modes converge to
the exact solution to 3 decimal places, whilst the third mode shape is only accurate to 2 decimal places.
305
Structural Engineering Analysis and Design
Second Edition
M
u1 u1 1 9
=
3m EI 12EI/33 12EI/33 12EI / 3 3 + 12EI / 3 3 8EI
M u2 u2
1 9
12EI/33 12EI/33 =
3m EI 12EI / 3 3 + 12EI / 3 3 8EI
M u3 u3
1 9
3m 2EI 12(2EI)/33 12(2EI)/33 =
12(2EI) / 3 + 12(2EI) / 3 16EI
3 3
The figure above shows a 3-degree of freedom shear frame. The flexibility formulation is described by
1
FMφ = φ 5 3 1 M 0 0
ω2 A = FM =
9
3 3 1 0 M 0
Ax = λx 16EI
1 1 1 0 0 M
4.1984M M
1
Sweeping matrix, S1 = I − φ1 φ1T M where Μ 1 = φ1T M φ1
Μ1
Second mode : Operate on A = FMS1
1 1.0000
= − 0.8,
0.5625M
= − 0.9994
(0) (12)
Guess x Ax
EI
− 0.8 − 0.9995
Convergence is achieved to 4 decimal places after 12 iterations. Thus, 1/ω 22 = 0.5625M / EI,
0.5625M M
1
Sweeping matrix, S 2 = S1 − φ 2 φ 2 M where Μ 2 = φ 2 M φ 2
T T
Μ2
Third mode : Operate on A = FMS 2
1 0.9999
= − 2,
0.3014M
Guess x (0)
Ax ( 3)
= − 2.7358
EI
3 3.7360
Convergence is achieved to 4 decimal places after 3 iterations. Thus, 1/ω 32 = 0.3014M / EI,
0.3014M M
306
Structural Engineering Analysis and Design
Second Edition
4.1.3.2.5 Natural (Fundamental) Frequency Using the Approximate Rayleigh Quotient Method
The stiffness formulated eigenvalue problem can be solved by employing the Rayleigh Quotient method. The
basis of the method is to equate the maximum kinetic energy during motion to the maximum potential energy to
obtain an estimate of the natural circular frequency with an assumed mode shape or displacement function. With
this method, only the fundamental natural circular frequency can be obtained fairly accurately. Vibration
analysis by Rayleigh’s Quotient method will yield a fundamental natural circular frequency ω1 which is higher than
the actual because of the approximation made by the mode shape φ(x). Higher natural circular frequencies can be
obtained only if the exact higher mode shapes are known.
(a) Guess mode shape based on the displacements produced by applying inertial forces as static loads. That is
to say, load the structure with the inertial loads associated with an assumed displacement. This load then
results in a new improved displacement configuration. Loop for even better estimate or simply goto (b).
(b) Apply Rayleigh’s formula to obtain an estimate of ωi
ωi2 =
Generalised Stiffness of Mode i, K i {φ } [K ]{φi } =
= iT
T
∑ k ∆φ
i
2
i
(c) From the obtained estimate of the eigenvalue, obtain a better estimate of the eigenvector from
[[K] − ωni2 [M]]{φ}i = 0
with the eigenvector normalized in the same method as before. Loop to (b) for a new estimate of the
eigenvalue based on the new eigenvector. This is repeated until convergence. The stiffness of the calculated
system is always greater than the stiffness of the exact system as we are modelling a system of infinite
DOFs with a finite number of DOFs. The value of ω1 decreases as the mode shape becomes closer to the
actual fundamental mode shape. Hence, the calculated natural circular frequency will always be greater
than the exact value. Seek to obtain the lowest possible estimate of ω for convergence.
M
u1 1.0
3m EI
M u2
0.8
3m EI
1.5M u3 0.7
3m 2EI
2M
u4 0.6
4m 4EI
307
Structural Engineering Analysis and Design
Second Edition
12 EI 12 EI 24 24
2 × 33 − 2×
33
0 0 27 − 0 0
27
24
− 2 × 12 EI 12 EI
4× 3 − 2×
12 EI
0 −
48
−
24
0
K= 33 3 33 = EI 27 27 27
12 EI 12 EI 12( 2EI) 12( 2EI) 24 8 48
0 − 2× 3 2× 3 + 2× − 2× 0 − −
3 3 33 3 3
27 3 27
12( 2EI) 12( 2 EI) 3(4 EI) 48 155
0 0 − 2× 2× + 2 × 0 0 −
33 33 4 3 27 72
M 0 0 0 1
0 0 0.93
M 0
M= φ1 =
0 0 1 .5 M 0 0.80
0.69
0 0 0 2M
ω12 =
K1 {φ }T [K ]{φ1 } = 0.2194EI = 0.0581 EI
= 1T ∴ ω1 = 0.2410
EI
M1 {φ1 } [M ]{φ1 } 3.7771M M M
The alternative approach to the calculation is to use the following formula. However, the former approach above is
more recommended as defining the storey stiffnesses below can be tricky.
ω12 =
∑ k ∆φ i
2
i
∑m φ i
2
i
12EI/33 12EI/33
12EI/33 12EI/33
12(2EI)/33 12(2EI)/33
3(4EI)/43 3(4EI)/43
Transverse stiffness
of columns
308
Structural Engineering Analysis and Design
Second Edition
The Rayleigh Quotient expression can also be rewritten in terms of the static displacements as follows. This is
employed in Earthquake Engineering codes of practice for the fundamental period of structures.
T1 = 2π
∑Wδ 2
i i
g∑ W δ
i i
The static deflections due to the horizontal application of the weight of the structure are obtained from a static
analysis program. It need not be the gravity forces that are applied, but any inertial force that produces a deflected
shape corresponding to the mode shape of interest. The static horizontal deflected shape of the structure resembles
the fundamental mode shape of the vibrational analysis because the p0i/Ki term is largest for the fundamental mode
since higher modes are less significant in contributing to the static response. This method however does not
employ the repetitive improved estimates of the Improved Rayleigh Quotient method until convergence of
the mode shape. Hence if the mode shape estimate is incorrect, so will be the natural frequency. Of course,
the static displacement should be calculated assuming a weight corresponding to the loading for which the
frequency is required, i.e. usually a dead load with an allowance for the expected live load. Vertical bending modes
of single span beams and plates can be found knowing the static deflections at various points (hence assuming
lumped mass DOFs at those points). For the horizontal mode of vibration of an entire structure, the gravity force
must be applied laterally. For multi-span beams, say a two span continuous beam, the fundamental mode is an
antisymmetric mode. Hence, if the static loads were applied all in the same direction, we would be predicting the
frequency of the symmetric mode, which is higher. This is an example where the assumed mode shape was totally
incorrect. The repetitive improved estimates of the Improved Rayleigh Quotient method above would have detected
this. The static load should have been applied downwards in one span and upwards in the other to produce a
deflected shape that somewhat resembles the fundamental asymmetric mode.
Lumped Mass Rayleigh Quotient Method – Lumped Mass & No Improvement To Guessed Mode Shape
The above Rayleigh Quotient expression can also be rewritten in terms of the static displacement and lumping the
mass to a single mass DOF. For SDOF systems, we have said that the natural frequency can be estimated from
1 k
fn =
2π m
This is the Rayleigh Quotient expression for SDOF systems if you like. Replacing the mass term with weight, W,
we have
g k g k g 1
fn = = = = 15.76 u st in mm
2π mg 2π W 2π u st u st
Which now relates the natural frequency to the static deflection ust of a single lumped mass dynamic DOF. This
formula is exact ONLY for single lumped mass systems. We can approximately extend this to MDOF systems
but the numerical factor 15.76 will be different generally between 16 and 20; 18 is quite practical. But of course,
this is a fudge to account for the fact that we are not taking static deflections at various mass DOFs, but
simply at one solitary mass DOF. The advantage of this method is that it is simple. If the static deflection is
known from various standard static deflection equations, the natural frequency can be determined, albeit very
approximately. Vertical bending modes of single span beams and plates can easily be found knowing the midpoint
static deflection (hence assuming a lumped mass DOF in the middle). For the horizontal mode of vibration of an
entire structure, the gravity force must be applied laterally and the lateral deflection at the top of the equivalent
vertical cantilever is employed (hence assuming that the lumped mass DOF is at the top of the building).
309
Structural Engineering Analysis and Design
Second Edition
The Rayleigh Quotient method gives the first mode of vibration only. Ritz extended this method to give a few more
modes.
But with the approximation of {φ} = [Ψ]{z}, the reduced eigenvalue problem can be expressed as
([K*] − ωi2[M*]) {zi} = {0} i = 1 to the number of Ritz vectors in reduced system, s < n
(b) The Ritz vectors [Ψ] = [{Ψ1},{Ψ2},…{Ψs}] are assembled, where s < n
(e) The s (not n) approximate mode shapes in the original system is obtained
(f) The number of good approximation of modes = s / 2; Hence, Ritz extended Rayleigh’s method to give
s / 2 modes with good accuracy
310
Structural Engineering Analysis and Design
Second Edition
4.1.3.3.1 Natural (Fundamental) Frequency Using the Approximate Rayleigh Quotient Method
The Rayleigh Quotient Method is approximate because of the approximation of the mode shape. The closer the
mode shape resembles the true mode shape, the more accurate the natural frequency estimation.
For the flexural Bernoulli-Euler beam of length L with additional lumped masses and stiffnesses the natural circular
frequency ωi is given by
n2
EI( x ){φ′′( x )} dx + ∑ k jφ2j
L
∫
2
Generalised Stiffness K i 0
j=1
ωi2 = = n1
Generalised Mass Mi
m( x ){φ( x )} dx + ∑ m i φi2
L
∫
2
0
i =1
The number of coefficients in the polynomial approximation of the mode shape φ(x)
= the number of geometric and static boundary conditions + 1
One is added because one coefficient will cancel out in Rayleigh’s formula.
311
Structural Engineering Analysis and Design
Second Edition
4.1.3.4 Natural Frequencies By Exactly Solving the Partial Differential Equilibrium Equation
In reality, all systems have distributed mass, damping and stiffness properties. The finite element method
discretizes the continuous system with a finite number of degrees of freedom resulting in a set of second order
ordinary differential equations (ODEs). Continuous systems on the other hand result in one partial differential
equation of at least second order. The reason for this is that on top of having the time variable, continuous systems
also include the spatial variable as there are not discrete DOFs. Because of the mathematical complexity of
generating and solving (integrating) these partial differential equations, this method is limited to simple structures.
The Bernoulli-Euler beam equation of motion, which is based on simple bending theory assuming that plane
sections remain plane, no shear deformation and no axial force effects results in the equation of motion
∂4u ∂2u
EI + m = P(x, t ) m = mass per unit length
∂x 4 ∂t 2
where u is the transverse displacement. Note that the Timoshenko beam equation extends the Bernoulli-Euler beam
theory to include shear deformations and rotary inertia and should be used if such effects matter. Solving the
corresponding free vibration PDE analytically by the method of separation of variables leads to two ODEs, which
yield the following natural circular frequencies and mode shapes
EI
ω n = (aL )
2
mL4
φ( x ) = A sin ax + B cos ax + C sinh ax + D cosh ax
By applying various boundary conditions to these equations, various solutions may be found. The modal mass is
L
Mi = m ∫0
φ( x ) 2 dx
Boundary Natural
(aiL)2 Term Mode Shapes Modal Mass
Conditions Frequencies
cos(aL)cosh(aL) – 1 = 0 φ i ( x ) = cosh a i x − cos a i x − σ i (sinh a i x − sin a i x )
EI
ωni = (a i L )2
Fixed-Free cosh a i L + cos a i L
thus (aiL)2 = 3.5160, 0.25mL
(Cantilever) mL4 σi =
22.0345, 61.6972 … sinh a i L + sin a i L
sin(aL) = 0
iπx
ωni = (a i L )
Simply EI aix
φi ( x ) = sin = sin
2
thus, (aiL)2 = (iπ)2, i = 1, 0.50mL
Supported mL4 L L
2, 3 …
Fixed-Simply
tan(aL) – tanh(aL) = 0 φi ( x ) = − cosh a i x + cos a i x + σi (sinh a i x − sin a i x ) M1 = 0.439mL
ωni = (a i L )
Support EI
thus (aiL)2 = 15.4118,
2
− cosh a i L + cos a i L M 2 = 0.437 mL
(Propped mL4 σi =
Cantilever)
49.9648, 104.2477 … − sinh a i L + sin a i L M3 = 0.437 mL
cos(aL)cosh(aL) – 1 = 0 φ i ( x ) = cosh a i x + cos a i x − σ i (sinh a i x + sin a i x )
ωni = (a i L )
EI
Free-Free thus (aiL)2 = 22.3733,
2
cosh a i L − cos a i L
mL4 σi =
61.6728, 120.9034 … sinh a i L − sin a i L
cos(aL)cosh(aL) – 1 = 0 φ i ( x ) = cosh a i x − cos a i x − σ i (sinh a i x − sin a i x ) M1 = 0.394mL
ωni = (a i L )
Fixed-Fixed EI
thus (aiL)2 = 22.3733, 2
− cosh a i L + cos a i L M 2 = 0.439mL
(Encastre) mL4 σi =
61.6728, 120.9034 … − sinh a i L + sin a i L M 3 = 0.437 mL
The modal mass for the simply supported beam is half the total mass for all the modes and that for the cantilever is
quarter the total mass for all modes.
312
Structural Engineering Analysis and Design
Second Edition
The natural frequencies of a simply-supported Bernoulli-Euler beam with axial force (positive P being tension) are
f ni =
(iπ)2 EI
1 +
PL2
2π mL4 EIi 2 π 2
313
Structural Engineering Analysis and Design
Second Edition
In reality, both shear and bending deformations must be accounted for when computing the frequencies of the
beam. Shear deformation become prominent in deep (with respect to the span) beams. Hence the equivalent
building cantilever beam model tends to be more of a shear mode. We are talking about the global behaviour of the
building and not the shear deformation of individual column members, which are generally flexural. Note that shear
frequencies are proportional to 1/L whilst bending frequencies are proportional to 1/L2.
The Southwell-Dunkerly formula can then be used to combine the effects of both bending and shear deformations
(of course both deformations have to occur in the same mode to be compatible) to obtain a frequency estimate as
follows.
1 1 1
2
= 2 + 2
f n f n bending f n shear
314
Structural Engineering Analysis and Design
Second Edition
For mb = 1 (higher modes in this direction of much higher frequencies), the natural frequencies are
For MAX normalized mode shapes, the modal masses for ALL modes of the simply supported anisotropic plate
are quarter of the total mass.
Flat Slab
Dx and Dy are flexural rigidity per unit width
For unit width beam, the flexural rigidity, D = EI = Eh3/12 per unit width whilst for unit width plate, the flexural
rigidity, D = EI = Eh3/[12(1-ν2)] per unit width. The torsional rigidity H should be such that it is equal to the
bending rigidity D so that there is equal bending stiffness of the plate about any horizontal axis. The natural
π Eh 3
frequency of a square flat slab plate simply supported all round is f =
a2 (
12m 1 − ν 2
.
)
315
Structural Engineering Analysis and Design
Second Edition
Dx is the flexural rigidity per unit width for bending about y-axis (or bending in x-z plane).
Dy is the flexural rigidity per unit width for bending about x-axis (or bending in y-z
plane); Iy = t.(H-h)3/12 + Aweb(lever arm)2 + (ay.h3/12)[ax/(ax-tortho+tortho.αortho3)] +
Aslab(lever arm)2; Note the factor [ax/(ax-tortho+tortho.αortho3)] to account for the effect of
the orthogonal ribs on the bending stiffness. Thus here αortho = h/Hortho where (Hortho -
h) is the depth of the orthogonal ribs and tortho is the thickness of the orthogonal ribs.
316
Structural Engineering Analysis and Design
Second Edition
A good approximate fundamental frequency formula for single-storey buildings, multi-storey buildings and towers
is (Bachmann and EC1-2.4:1995)
f = 46 / h (Hz)
where h is the height (m).
where
Another approximations is
where
A good approximate fundamental frequency formula for cable stayed bridges is (Bachmann)
317
Structural Engineering Analysis and Design
Second Edition
f = 110 / L
where L is the length of the main span (m).
318
Structural Engineering Analysis and Design
Second Edition
{u(t)}i = {φ} Re al [e
i
αt
i
[((ξ1Ri + ξ 2 R ) cos ωdi t − (ξ1Ii − ξ 2 Ii ) sin ω di t ) + i((ξ1Ii + ξ 2 Ii ) cos ωdi t + (ξ1Ri − ξ 2 Ri ) sin ω di t )]]
For the free vibration response to be real for all t,
ξ1Ii = −ξ 2 Ii and ξ1Ri = ξ 2 Ri
Since there are two less independent constants, let
ξ1Ii = −ξ 2 Ii = ξ Ii and ξ1Ri = ξ 2 Ri = ξ Ri
We notice that ξ1i and ξ 2i are a complex conjugate pair
ξ1i = ξ Ri + iξ Ii and ξ 2i = ξ Ri − iξ Ii
Hence
{u(t)}i = {φ}i e αi t (2ξ Ri cos ω di t − 2ξ Ii sin ω di t )
Thus the total response due to the superposition of all modes
{
{u(t)} = [Φ ] e αi t (2ξ Ri cos ω di t − 2ξ Ii sin ω di t ) }
In the real eigenvalue analysis, for each mode i, NASTRAN outputs,
one real root ω ni representing the undamped natural circular frequency
one real eigenvector {φ}i representing the real mode shape
Complex eigenvalue analysis with viscous (and structural optional) damping, for each mode i, NASTRAN outputs,
two complex roots which are a complex conjugate pair, α i + iω di and α i − iω di , where
α i is negative and represents the decaying constant
ω di is positive and represents the damped natural circular frequency
two complex eigenvectors which are a complex conjugate pair, {φ R + iφ I }i and {φ R − iφ I }i
Complex eigenvalue analysis with only structural (and no viscous) damping, for each mode i, NASTRAN outputs,
one complex root, α i + iω di , where
α i is negative and represents the decaying constant
ω di is positive and represents the damped natural circular frequency
one complex eigenvector, {φ R + iφ I }i
319
Structural Engineering Analysis and Design
Second Edition
A pure real mode has all its terms in phase which each other. Different modes will of course be out of phase with
each other, but every point in a particular mode will vibrate in phase. And so, for a mode, all points in the structure
reach its maximum and minimum at the same phase instant. A complex mode on the other hand will have different
points of the structure reaching its maximum at different phase instants, i.e. different points of the structure are out
of phase with each other for a particular mode. Of course, different modes are still out of phase with each other.
The degree to which these points are out of phase with each other is a measure of the complexity of the mode. An
Argand diagram with all the terms of the complex eigenvector plotted will exhibit a narrowband if the mode is only
slightly complex. The diagram below shows plots for 4 eigenvector terms. Since there is only a slight phase
difference between the terms (angle between the arrows), the 4 terms define an almost real complex eigenvector.
Im
Re
The following eigenvector plot on the other hand is highly complex because of the large phase angle difference
between the terms of the eigenvector.
Im
Re
The animation of the complex modes is quite indicative of the complexity of the mode. In complex modal analysis,
each mode has two complex conjugate eigenvectors. Each eigenvector has a real and imaginary part. The animation
of such a complex mode is only of value when we plot the magnitude of either (of the two complex conjugate)
eigenvectors. The phase information (which is obtained when the magnitude and argument of the real and
imaginary components are obtained) is essential in order to determine the relative phase of the motion of different
parts of the structure. In a real mode, there is no phase information, and hence the animation of a real mode will
show all parts of the structure vibrating in-phase with each other. Explicit viscous dampers will cause parts of the
structure to vibrate clearly out-of-phase. This clearly occurs when explicit viscous dampers are used to damp
prestressed cable vibrations. A cable with a damper element of a very high damping coefficient attached to its
center will behave as two separate cables, the damper element effectively producing a fixity at the center of the
cable. This changes the fundamental mode shape of the cable from one of a low frequency to two cable mode
shapes of higher fundamental frequency.
Because the solution by complex modal analysis yields a decaying constant αi for each and every mode, the modal
damping can be obtained as modal structural damping, Gi = 2αi/ωdi.
320
Structural Engineering Analysis and Design
Second Edition
SOL 107
$ CASE CONTROL SECTION
$ Sets defining grid ids or element ids
SET < Number > = 1 THRU 100, 211, 343, < etc >
$ Grid output of displacement for each mode i.e. eigenvector
$ SORT1 lists the results by eigenvalue whilst SORT2 lists the results by grid id
DISPLACEMENT(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
$ Grid output of displacement for each normalized mode i.e. normalized eigenvector
SDISPLACEMENT(<SORT1/SORT2>,<PRINT,PUNCH>,<REAL/PHASE>) = ALL/<Grid Set ID>
$ Grid output of real eigenvector for the a-set
SVECTOR(<PRINT,PUNCH>) = ALL/<Grid Set ID>
$ Grid output of SPC forces
SPCFORCES(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
$ Element output of force, stress and strain
ELFORCE(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Element Set ID>
ELSTRESS(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Element Set ID>
STRAIN(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Element Set ID>
$ Analysis Requests
SPC = < ID of SPC Cards Defined in Bulk Data >
CMETHOD = < ID IN EIGC >
$ XY plot output
OUTPUT(XYPLOT)
XYPUNCH <DISP/VELO/ACCE> RESPONSE <subcase>/<Grid ID>(<T1/T2/T3><RM/IP>)
XYPUNCH <ELFORCE/ELSTRESS/STRAIN> RESPONSE <subcase>/<Element ID>(<Code Number>)
The METHOD commonly used is CLAN. NORM, G, and C have to do with the specification of the method of
normalization. The eigenvectors may be normalized either to a unit value at grid point G for coordinate C, or for
the largest term to be of unit magnitude. E is used to specify the convergence criterion of the solution. Each method
has a different default value for this criterion, and each is adequate for most problems. The following continuation
is repeated for each desired search region. (J = 1 to n, where n is the number of search regions).
As mentioned, replacing the following into the free damped equation of motion
where
321
Structural Engineering Analysis and Design
Second Edition
If α=0, the radian frequency, ω of complex eigenvalue analysis is the same as that of real eigensolutions. The real
part, α, is a measure of the decay rate of a damped structure, or if negative, the rate of divergence of an unstable
system. The imaginary part, ω, is the modified frequency in radians/unit time. However, roots with negative values
of ω should be treated as special terms. The recommended practice is to specify one point α1=0.0 and ω1 at the
lower bound of the expected range of eigenvalues, but not at 0.0. A second shift may be input at α2=0.0 and ω2 at
the upper bound of the expected range. All ALPHABj and OMEGABj must be blank.
The column labeled (REAL) contains α1, and the column labeled (IMAG) contains ω1. The column labeled
(FREQUENCY) contains the circular frequency. The last column is the damping coefficient computed from the
equation
which is approximately twice the value of the conventional modal damping ratio. Note that if the magnitude of this
term is computed to be less than 5.0x10-4, it is reset to zero. For small values, the damping coefficient is twice the
fraction of critical damping for the mode. The eigenvalues are sorted on ω, with the negative values sorted first
(there are none in this example), sorted on increasing magnitude, followed by the eigenvalues with positive ω,
again sorted on magnitude. Roots with equal ω values are sorted next on α.
322
Structural Engineering Analysis and Design
Second Edition
It is often necessary to incorporate the reduction in bending stiffness of gravity load resisting columns for the
analysis of lateral loads. The following procedure is undertaken.
Phase 1
Perform static analysis (with loads that cause the greatest negative or positive geometric stiffness) based on [KEA]
but include the segyroa.v2001 alter prior to the Case Control Section in order to compute the geometric stiffness
matrix [KGA]1 (and output into a .pch file) based on the generated element loads from the [KEA] static analysis.
Phase 2
A SOL 107 is undertaken based on [KEA] + [KGA]1 with the k2gg=ktjj statement in the Case Control Section and the
outputted .pch file which contains the ktjj matrix in the Bulk Data. The responses at this stage represent the P-∆
(KGA From KEA) complex modal response.
The method is valid when only the prestress is judged to affect the geometric stiffness such as in the
compressive preload of building columns due to gravitational loads and the prestressing of extremely taut cables
that sag very little under gravity but not in systems such as suspension bridges. Where lateral loads are large
enough to affect the geometry of the system with prestress, then a repetitive P-∆, SOL 106, implicit dynamic
relaxation SOL 129 or explicit dynamic relaxation must be employed. But in single P-∆ analysis, because cables do
not have much elastic bending stiffness, the initial static preload subcase should only include the prestress and not
gravity as including gravity is the same as solving two linear static problems of stiffness KEA with preload and
gravity as the applied loads respectively. Clearly, in the gravity case, it is nonsensical as the cables do in reality
have differential stiffness (from the prestress) to resist the gravitational force. Prestress in one direction (i.e. along
the axis of cable) will cause a differential stiffness in the orthogonal direction. Gravity acts in the orthogonal
direction and hence cannot be accounted for in the calculation of the prestress in this single P-∆ analysis. To
quantitatively decide if gravity need not be considered in contributing to the differential stiffness of the cables, a
static P-∆ analysis should be carried out, the first subcase being a SOL 101 with only the prestress as applied loads
and the second subcase a P-∆ SOL 101 (i.e. utilizing the induced prestress from the first subcase to form a
geometric stiffness matrix) with both the gravity and prestress included as applied loads. If the difference in the
cable element forces between subcases 1 and 2 is negligible, then gravity has little influence in affecting the
geometric stiffness. If there is a major difference in the cable element force, then clearly, gravity will affect the
geometric stiffness and as such, a repetitive P-∆ , SOL 106, implicit dynamic relaxation or explicit dynamic
relaxation must be used to converge to the true KT. Likewise, in the single P-∆ analysis of multi-storey buildings,
gravity (and only gravity) acts in axis of columns to generate prestress, and the differential stiffness is computed for
the orthogonal direction reducing resistance to lateral wind forces, applied in the second subcase with gravity too.
323
Structural Engineering Analysis and Design
Second Edition
The STATSUB(PRELOAD) computes the differential stiffness due to the prestress and also the follower force. The
follower force is calculated and incorporated by the use of PARAM, FOLLOWK, YES. We know how the prestress
affects the differential stiffness, namely a tensile prestress causing an increase in stiffness. The effect of the follower
force on the stiffness is different. For example, for a cylinder under external pressure critical buckling load may be over-
estimated (even though the mode shapes are similar) in a SOL 105 and the natural frequencies in vibration may be under-
estimated (even though the mode shapes are similar) in a SOL 103 in the absence of follower stiffness. To the contrary,
these observations are reversed in case of centrifugal loads. Centrifugal forces as a constant (static) load are applied by a
Bulk Data RFORCE to any elements that have masses. The follower stiffness due to centrifugal load has the effect of
lowering stiffness (although the centrifugal load tensioning effect increases stiffness), consequently lowering natural
frequencies (even though the mode shapes are similar) in a SOL 103 and lowering the buckling loads (even though the
mode shapes are similar) in a SOL 105. This effect increases as the RPM increases, and it becomes significant when the
RPM is over 1000. For moderately geometric nonlinear analysis, exclusion of follower stiffness affects the rate of
convergence, but the converged solution is correct. For severely geometric nonlinear analysis, it may not be
possible to obtain a converged solution without including follower stiffness. As the geometric nonlinearity
intensifies, so is the effect of follower stiffness. Therefore, inclusion of follower stiffness greatly enhances the
convergence if the deformation involves severe geometric nonlinearity.
324
Structural Engineering Analysis and Design
Second Edition
4.2.4.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Complex Modal Analysis
The above top deck solves the linear modal complex eigenvalue problem. It is often necessary to include the
differential stiffness, especially if there are prestressed cables in the model and the mode shapes are sought. To
obtain KTA, to be theoretically exact, a GNL SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic
relaxation LS-DYNA) with prestress (as temperature loads say) and gravity must be undertaken. Alternatively, an
approximation to KTA can be obtained by repetitive P-∆ static analyses with the prestress (as temperature loads say)
and gravity applied. The procedure to obtain this approximate KTA will be presented. Note that the approximate KTA
will be the summation of the elastic stiffness KE at the undeflected (by the prestress and gravity) state but KG at the
deflected (by the prestress and gravity) state. Hence if KE changes considerably during the application of the
prestress, a full SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-DYNA),
which converges to the KE and KG at the deflected (by the prestress and gravity) state should be employed. Hence
for the modelling of a suspension bridge where there is a great change in geometry (known in the bridge industry as
form-finding, so-called because it is necessary to find the form or shape of the catenary suspension cables), it may
be prudent to employ SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-
DYNA), but for a high tension low sag cable on say a tower with prestressed cables, the repetitive P-∆ static
analysis may be adequate. The repetitive P-∆ analysis basically involves a number of iterations of linear static
analyses to obtain an approximate KTA. Note again that A refers to the initial undeflected (by the collapsing load)
state, but deflected by the prestress and gravity. To perform the repetitive P-∆ analysis, a static analysis is
performed based on KEA with temperature loads and gravity to generate forces in the structural elements, which in
turn provides input for the computation of KGiAKTm where m is the iterations. Repetitive static analysis is performed
with the prestress and gravity updating the stiffness matrix KEA + KGiAKTm-1 + KGiAKTm until convergence of
displacements is obtained. The tangent stiffness at this stage is the approximate converged tangent stiffness matrix
KTA = KEA + KGiAKT. The converged displacements represent the approximate P-∆ (KGA From Approximate KTA)
static response to the initial prestress loads. The converged geometric stiffness at this stage would be that based
upon the approximate tangent stiffness matrix KTA, i.e. KGiAKT.
Phase 1
Perform static analysis (with prestress and gravity) based on [KEA] but include the segyroa.v2001 alter prior to the
Case Control Section in order to compute the geometric stiffness matrix [KGA]1 (and output into a .pch file) based
on the generated element loads from the [KEA] static analysis.
Phase 2
Perform static analysis (with prestress and gravity) based on [KEA] + [KGA]1 by including the k2gg = ktjj statement
in the Case Control Section, the outputted .pch file which contains the ktjj matrix in the Bulk Data and the
segyroa.v2001 alter prior to the Case Control Section to compute the [KGA]2 (and output into the .pch file
overwriting previous data) based on the generated element loads from the [KEA] + [KGA]1 static analysis.
Phase 3
Repeatedly perform the Phase 2 static analysis (with prestress and gravity) based on [KEA] + [KGA]i for i = 2 to n
where n represents the number of iterations required for the change in deflections between analyses to become
negligible. This would signify that the change in the [KGA] matrix become negligible and the correct [KGA] is
attained. The deflections and the other responses at this stage represent the P-∆ (KGA From Approximate KTA) static
response to the prestress and gravity. The stiffness of the structure is KTA.
Phase 4
A SOL 107 is undertaken with the k2gg=ktjj statement in the Case Control Section and the outputted .pch file
which contains the latest ktjj matrix in the Bulk Data. The responses at this stage represent the P-∆ (KGA From
Approximate KTA) complex modal response.
325
Structural Engineering Analysis and Design
Second Edition
4.2.5.1 Determination of Damped Natural Frequency and the Maximum Dynamic Displacement, umax for
Free Damped Vibration Due to Initial Displacement and/or Initial Velocity by Classically Solving
the SDOF Linear ODE and Maximizing the Solution
Case : critically damped c = c cr = 2 km = 2mω n , the two roots are real and repeated,
c cr 2 km
λ1 = λ 2 = − =−
2m 2m
Hence,
c cr
−
u(t) = (G 1 + G 2 t )e
t
2m
Case : overdamped c > c cr , the two roots will be real and distinct,
u(t) = G 1e λ1t + G 2 e λ 2 t
Again, there will be no oscillatory motion, instead exponential decay in a longer time.
326
Structural Engineering Analysis and Design
Second Edition
−ct
u ( t ) = Re al e 2 m [((G 1R + G 2 R )cos ωd t − (G 1I − G 2 I )sin ωd t ) + i((G 1I + G 2 I )cos ω d t + (G 1R − G 2 R )sin ω d t )]
for the free vibration response to be real for all t,
G 1I = −G 2 I and G 1R = G 2 R
sin ce there are two less independen t constants, let
G 1I = −G 2 I = G I and G 1R = G 2 R = G R
we notice that G 1 and G 2 are a complex conjugate pair
G 1 = G R + iG I and G 2 = G R − iG I
hence
c
−
[(2G R cos ωd t − 2G I sin ωd t )]
t
u(t) = e 2m
ωd =
k ζ 2 km
−
m 2m
=
k ζ2k
−
m m
=
k
m
( )
1 − ζ 2 = ωn 1 − ζ 2
hence,
u ( t ) = e −ζωn t [(2G R cos ωd t − 2G I sin ωd t )]
where ωd = ω n 1 − ζ 2 and 2G R , − 2G I being the constants of integratio n from the initial conditions .
Note also that we often define the decaying constant
c
α=−
2m
hence since c = ζ 2mω n
ζ 2 mω n
α=−
2m
α = −ζω n
327
Structural Engineering Analysis and Design
Second Edition
Damping causes the damped natural frequencies to be a little less than the undamped natural frequencies, but the
difference is negligible in real structures, which are under-damped. Hence, the old classroom example of measuring
the natural frequency of a pendulum in air and in water gives a lower frequency in water simply because of the
added water mass in the dynamic model.
The logarithmic decrement is defined as the natural log of the ratio between successive peaks (m = 1),
un 2πζ
δ = ln = ≈ 2πζ
u n +1 1− ζ2
328
Structural Engineering Analysis and Design
Second Edition
The solution method can be used to solve dynamic systems subjected to: -
(a) Deterministic periodic harmonic long duration loading functions
A deterministic periodic forcing function has regularly repeating amplitude. The sine or cosine function is said to
be harmonic. Because the periodic function repeats itself, any initial starting transient response is insignificant and
the steady state response is of interest, hence the solution is performed in the frequency domain. The starting
transient normally decays away after 50-100 cycles of oscillation for light damping. The steady-state oscillatory
response occurs at the same frequency as the loading phase shifted due to damping.
In this LINEAR FREQUENCY DOMAIN solution, not only that the static response has to be added separately,
but also the mean of the dynamic excitation has also got to be added separately as a static response. This is because
the mean of the dynamically applied force is not included in the dynamic excitations. Hence the total response in
this frequency domain dynamic analysis = static response to mean of dynamic excitation + dynamic response
+ static response to static loads.
In the modal frequency analysis, the solution of a particular forcing frequency is obtained through the summation
of the individual modal responses. With the knowledge of the modal frequencies and corresponding mode shapes,
the coupled system of simultaneous dynamic equilibrium equations can be uncoupled by premultiplying the
dynamic equation of motion by [Φ]T as shown below. This means that the simultaneous ODEs no longer need to be
solved simultaneously and so the MDOF system of equations is reduced to a SDOF system of equations, which can
be solved independently of each other. The computational benefit comes from not having to run a rigorous
simultaneous equation solver. The knowledge of the modal frequencies and the corresponding mode shapes are a
prerequisite to this implicit modal frequency response analysis, hence the implicit real (note not complex)
eigenvalue analysis must be performed first. Because of the fact that the modes are real and do not take into
account the explicit elemental (structural and viscous) damping within the structure, the coupled ODEs cannot be
uncoupled by [Φ]T. The system of coupled ODEs remains in the existence of structural and/or viscous damping and
the simultaneous equation solver must be invoked. The only difference then between SOL 111 and SOL 108 is the
fact that the former solves the equations in the modal coordinates instead of the physical coordinates. Hence, as
long as sufficient modes are included (to represent both the static and dynamic response of the system), SOL 111
will give the same solution as SOL 108 even with large values of structural and/or viscous damping. We know that
large values of elemental damping will modify the mode shape and frequencies. However, a SOL 111 can still be
performed because a true (real) modal approach is not undertaken, instead the coupled ODEs are solved using a
direct approach, but in the modal coordinates. However, the use of the FREQ4 card, which bases the excitation
frequencies to be solved for on the real natural frequencies may prove to be insufficient. This is because with high
values of elemental viscous damping, certain local modes can be totally eliminated. For instance, viscous dampers
with high coefficients of damping on cables can considerably alter the natural frequency of the local mode and even
eliminate a local mode altogether. In this case, the FREQ4 card will not capture the response at the damped natural
frequency. Hence, these damped natural frequencies need to be known (by performing a SOL 107) before choosing
the excitation frequencies to be solved for using FREQ2 cards. To reiterate, SOL 111 reverts to a direct solution
technique (akin to SOL 108) when there is either viscous or structural element damping as the equations of motion
cannot be orthogonalized. But the unknowns are still the modal responses and not the physical responses. And
because with damping the modes are really complex modes, there is greater difficulty in capturing all the response
using the real modes. Hence, even more modes are thus required to capture the response accurately in SOL 111
with element damping especially if the elemental damping is high enough to cause significant complex modes.
Thus if elemental damping is high and the modes are complex, the direct method SOL 108 may be more
appropriate. If a true modal approach is intended for the solution of coupled ODEs with structural and/or viscous
damping, then a complex modal forced frequency response analysis is necessary, although this is quite impractical.
329
Structural Engineering Analysis and Design
Second Edition
Frequency domain loading function from the time domain loading function
[
{P( t )} = Re al {P(ω)}e iωt ]
where {P(ω)} is the complex loading function vector
Frequency domain complex modal response to {P(ω)} is
{ξ(ω)} which is complex in general
Frequency domain complex total response function vector
{F(ω)} = [Φ ]{ξ(ω)}
where {F(ω)} is complex in general
Time domain total response from the frequency domain complex total response
[
{u ( t )} = Re al {F(ω)}e iωt ]
330
Structural Engineering Analysis and Design
Second Edition
The matrices [M] and [K] are diagonal because of the orthogonality condition of the normal modes
{φi }T [M ]{φ j } = 0 and {φi }T [K ]{φ j } = 0 if i≠ j
Accordingly, the generalised mass solitary term and the generalised stiffness solitary term
associated with a particular mode ' i' are
M i = {φ i } [M ]{φ i } K i = {φ i } [K ]{φ i } = ω 2ni M i
T T
and
And because these matrices are diagonal, the equations become uncoupled and is easy to solve for {ξ(ω)}.
Hence, the set of uncoupled SDOF systems is
−ω 2 M i ξ i (ω) + K i ξ i (ω) = Pi (ω)
Pi (ω)
ξ i (ω) = for i = 1 to n modes
− ω 2Mi + K i
which is solved for the individual modal responses ξ i (ω).
Finally, the total physical response is the superposition of the individual modal responses
{u ( t )} = Re al [[Φ ]{ξ(ω)}e iωt ]
{u ( t )} = Re al [{F(ω)}e iωt ]
In modal frequency response analysis, mode truncation refers to not utilizing all the natural modes in computing the
response of the structure. At a minimum all the modes that have resonant frequencies that lie within the range of
the forcing frequencies have to be retained. For better accuracy, all the modes up to at least 2 to 3 times the highest
forcing frequency should be retained. For example, if a structure is excited between 200 and 2000 Hz, all modes
from 0 to at least 4000 Hz should be retained as the forcing frequencies still excite the higher modes although not
resonating with them, i.e. off-resonant excitation. The recommendation of using modes up to 2 or 3 times the
highest excitation frequency assumes that the static response can be captured adequately using this finite number of
modes. Clearly, if the distribution of applied loads is multiple concentrated forces, then the finite number of modes
may still not be sufficient to capture the difficult static shape. In this case, it may be prudent to use static residual
vectors instead of increasing the number of vibration modes solely to capture the static response.
331
Structural Engineering Analysis and Design
Second Edition
If explicit viscous damping element contributions are made to the damping matrix [C] or if elemental structural
damping is specified such that a complex stiffness matrix exists, then neither the damping nor the stiffness matrices
can be diagonalized by the modal matrix. In this case the modal frequency response analysis solves the still coupled
problem using a direct frequency approach but in terms of the modal coordinates {ξ(ω)} instead of the physical
coordinates {u(t)}. Since the number of modes used in a solution is typically much less than the number of physical
variables, using the coupled solution of the modal equations is less costly than using physical variables. Hence,
with elemental viscous and/or elemental structural damping, the coupled system of ODEs of the damped
harmonically forced vibration equation of motion is given by
[M ]{&
u&( t )} + [C ]{u&( t )} + [ K ]{u ( t )} = {P ( t )}
where
[ K ] = (1 + iG )[ K ] + i ∑G E [K E ]
[C ] = ∑ [C E ]
Hence,
[− ω [Φ ] [M ][Φ ] + iω[Φ ] [C][Φ ] + [Φ ] [K ][Φ ]]{ξ(ω)} = [Φ ] {P(ω)}
2 T T T T
The modal frequency response analysis solves the still coupled problem using a direct frequency approach but in
terms of the modal coordinates {ξ(ω)} instead of the physical coordinates {u(t)}.
If however only viscous modal damping ζi is specified, the generalized damping matrix [Φ]T[C][Φ] remains
diagonal and so the uncoupled equations of motion can be maintained. The uncoupled equations of motion become
−ω2Mi ξ i (ω) + i(ζ i 2Mi ω n i )ωξ i (ω) + K i ξ i (ω) = Pi (ω) for i = 1 to n modes
Pi (ω)
ξ i (ω) =
− ω2Mi + i(ζ i 2M i ωn i )ω + K i
which is solved for the individual modal responses ξ i (ω).
The animation of the forced frequency response is quite indicative of the nature of the response of the structure to
harmonic excitations. The forced frequency response analysis will yield for each node, both real and imaginary
responses. The animation of the forced response of the structure as a whole (at any particular frequency of
excitation) is only of value when we plot the magnitude of the response, and not the individual real or imaginary
components. The phase information (which is obtained when the magnitude and argument of the real and
imaginary components are obtained) is essential in order to determine the relative phase of the motion of different
parts of the structure. Explicit viscous dampers will cause parts of the structure to vibrate clearly out-of-phase.
332
Structural Engineering Analysis and Design
Second Edition
If only structural modal damping Gi is specified, modal damping is processed as a complex stiffness of each
individual mode and so the uncoupled equations of motion can still be maintained. The uncoupled equations of
motion become
−ω2Mi ξ i (ω) + (1 + iG i (ω)) K i ξ i (ω) = Pi (ω) for i = 1 to n modes
Pi (ω)
ξ i (ω) =
− ω2M i + (1 + iG i (ω)) K i
which is solved for the individual modal responses ξ i (ω).
It is important to know how to convert between structural damping and equivalent viscous damping for a particular
mode. For a particular mode,
ζ i (2MiK i / Mi ) = G iK i
2ζ i = G i
Hence, the relationship Gi = 2ζi can be used to specify an equivalent viscous damping for structural damping, but is
only exact when the forcing frequency ω = the natural modal frequency ωni which occurs at resonance, i.e. when
damping is of primary importance.
{F(ω)} = {P(ω)}
−ω 2
[M ] + iω[C] + [K ]
The complex transfer function is defined as the complex frequency response function due to unit harmonic
excitations
[H(ω)] = 1
− ω [M ] + iω[C] + [K ]
2
This is the so-called transfer function that transfers the excitation to the response as follows
[
{u ( t )} = Re al [H (ω)]{P(ω)}e iωt ]
The (magnitude of the) dynamic amplification function D(ω) is defined as the magnitude of the complex response
function {F(ω)} divided by the static displacement.
333
Structural Engineering Analysis and Design
Second Edition
4.3.3 Capability of A Finite Number of Modes To Model The Static and Dynamic Response of Structure
When running a modal based dynamic response analysis, it is imperative to know if the chosen number of modes
within the modal dynamic analysis is capable of representing both the static and dynamic response of the structure.
First and foremost, the chosen modes must cover and extend beyond the excitation frequency range such
that both the resonant response of the modes within the range and the off-resonant response of the modes
beyond (from say a 1/3rd of lowest excitation frequency to 3x the highest excitation frequency) the range are
captured. This will sufficiently model the dynamic response. Considerations then must be made for the finite
number of modes to capture the static response.
It is often said in dynamic analyses that if the chosen finite number of modes are capable of representing the static
response, then they will usually also be able to represent the dynamic response of the structure. We shall consider
the frequency domain as this enables us to plot the FRF (and hence of course also the transfer function and the
dynamic amplification function). Let us picture the (magnitude of the) dynamic magnification function, here shown
below for a single mode, i.e. a SDOF system.
Damping Controlled
Dmax ≈ Dresonant = 1/(2ζ)
Stiffness Controlled
D Inertia Controlled
ω
0
ωn
The finite number of modes should be able to represent the true static displacement at zero excitation frequency and
the maximum (magnitude of the) dynamic amplification. Obtaining just the correct maximum (magnitude of the)
dynamic amplification is not proof that the chosen finite number of modes is sufficient. This is because the
response D(ω) at the frequency corresponding to the maximum dynamic response is controlled only by the
damping within the structure. Note that the corresponding value F(ω) is also a function of the modal stiffness (or
modal mass), the natural frequency and the (amplitude of the) modal force. Hence if the static response is
inaccurate, so will be the dynamic response. Thus it is best to compare the response function F(ω) between the
modal model and the full MDOF model.
To compare individual components of the modal model, the following approach is undertaken. First, the accuracy
of the stiffness representation is investigated. This corresponds to the static displacement. When the excitation
frequency is zero, if the modal approach with SOL 111 achieves the same F(ω) response as that from a static
analysis SOL 101, then the chosen finite modes are capable of representing the stiffness within the system and
hence are capable of modelling the static response for the system. The mathematics of converting the static
equilibrium equations in physical coordinates to equations in modal coordinates (and hence with a certain degree of
modal truncation unless the number of modal coordinates equals the number of physical coordinates) is presented
in Section 3.1.4.
Thereafter, the damping representation is evaluated by comparing the response F(ω) at the frequency of maximum
amplification. Finally, the inertial representation is evaluated by comparing the response F(ω) at high excitation
frequencies. If the responses obtained from the modal approach when the excitation frequency is off-resonant is
similar to that obtained from the multi degree of freedom model, then the chosen finite number of modes are
capable also of modelling the mass and stiffness distribution within the system.
334
Structural Engineering Analysis and Design
Second Edition
Hence, in order to quantitatively decide whether the chosen number of finite modes is capable of representing a
multi degree of freedom system the response between a direct solution SOL 108 and the modal solution SOL 111 is
compared. The complex frequency response function F(ω) can be plotted for the response at the DOF of interest
from both a SOL 108 and a SOL 111. A XYPLOT showing the contribution of individual modal responses F(ω)
superposed on a plot containing the total physical response is most useful in visually determining the contribution
of each and every mode to the static and dynamic response at particular frequencies of excitation. In
MSC.NASTRAN, it is possible to plot the response function F(ω) for individual natural modes using SOL 111
(even if there is structural and/or viscous damping as SOL 111 then performs a coupled ODE solution process, but
with the unknowns as the modal responses instead of the physical responses) by delimiting the mode calculated
using EIGRL or using LFREQ (lower limit of the frequency range), HFREQ (upper limit of the frequency range) or
LMODES (number of lowest modes retained). Alternatively, selective modes which are not sequential can be
deleted by the user by using the alter delmodea.v2001 and the associated bulk data entry DMI. The degree of
similarity between the two traces is an indication of the ability of the finite number of modes to represent the
response of the system. The number of modes should be increased or decreased accordingly. Note that the static
response at zero excitation frequency obtained from both SOL 111 and SOL 108 can be compared with the true
static solution from a SOL 101. It will be found that SOL 108 will yield the same static response as SOL 101 (when
there is no structural damping within the finite element model) whilst SOL 111 will yield a result that is lower due
to the nature of the finite number of modes presenting a stiffer response. Incidentally, since SOL 108 does not
allow the automatic specification of the frequency points to correspond to the natural frequencies of the structure,
the user would have to perform a SOL 103 and manually specify to increase the resolution near the vicinity of the
natural frequencies.
It is essential that the sufficient modes be incorporated when derived quantities are intended. For instance, to
evaluate the modal damping of a particular mode, both the static response and the dynamic response must be
accurately represented. If the static response is not sufficiently captured, then the dynamic response will not be
accurate either as the response of the FRF = P/K times D(ω). If the distribution of the excitation force is
concentrated at multiple points, then it is likely that the difficult static shape requires many modes of vibration. In
this case, it may be better to utilize static residual vectors to capture the response more directly. Alternatively of
course, the direct frequency approach SOL 108 may be used, in which case the static response and the dynamic
response of all modes are automatically captured, but it requires effort in capturing the peak response (by
requesting fine frequency intervals near the peaks). But when using the modal method, damping estimates using the
half power bandwidth on the FRF of the D(ω) or using 1/(2ζ) on the D(ω) depend on the capability of the modes to
accurately represent the static and dynamic response. Where in doubt, use a SOL 101 solution to check the static
response. And a SOL 107 solution should be used to check the modal damping estimates.
The relative contribution of each mode to the total static and dynamic response can also be determined. This is
known as the modal responses, i.e. the value of the modal variables in the modal solution scheme. The Case
Control Cards SDISP, SVELO and SACCE will produce the modal solution set output, {ξi(ω)}. These are plots of
the modal response (for each mode) in terms of frequency of excitation, i.e. the frequency domain modal response.
That is to say, SDISP(PUNCH) (or SVELO or SACCE) will output one value for each mode at each excitation
frequency. These are the modal responses in modal space. They must be multiplied by the corresponding mode
shapes in order to obtain the modal responses in physical space i.e.
{Fi(ω)} = {φi}ξi(ω)
Another method of determining the modal contribution of individual modes to a particular excitation is by using the
modal strain energy. Firstly, to evaluate the ability of the natural modes to represent the static response, a SOL 101
is run with the amplitude of the dynamic loads applied as a static load and the results are written to a DMIG file
using the alter pchdispa.v2001. Then a SOL 103 us run with the DMIG included and also the alter modevala.v2001
to ascertain how well each and every mode can represent the static solution. The strain energy for each mode can
be compared to the strain energy in all the modes calculated and also the input vector. Note that the eigenvector
scaling must be set to the default MASS (not MAX) for this alter to be valid. Secondly, to evaluate the ability of the
natural modes to represent the dynamic response, the alter mfreqa.v2001 is utilized for SOL 111. The results are
presented in a matrix of strain energies with the rows representing the excitation frequencies and the columns
335
Structural Engineering Analysis and Design
Second Edition
representing the natural modes. The frequency interval between the rows and between the columns corresponds to
the excitation frequency interval and the natural mode frequency interval. Of course, the dominant modes that
represent the static response is not necessarily the same as the dominant modes that represent the dynamic response
at different excitation frequencies. This is simply because the dynamic response is also a function of the loading
frequency and resonance frequency, on top of the static response. Including the dominant modes that represent the
static response will however improve the accuracy of representing the dynamic response F(ω) as it is the function
of the amplification D(ω) and the static response P/K.
Note that the capability of a finite number of modes to represent the static and dynamic response of the structure is
limited to a particular force excitation direction and distribution. If the force changes its location or direction,
then the prominence of the different modes which will be different. This is simply because changing the location or
direction of the force will cause a different set of modal forces and hence a different level of excitation of the
different modes. The capability of a finite number of modes to represent the static and dynamic response of the
structure depends also on the level of elemental structural and elemental viscous damping present within the
system. This is because, the level of elemental damping affects the complexity of the modes, hence possibly
requiring more real modes if the modal complexity is high. Thus damping should be included when performing the
study to determine the required number of modes.
For accurate results of modal methods, static residual vectors (PARAM, RESVEC, YES) can be calculated
except when using the seismic large mass method for enforced motion. The residual vector method is more
efficient than the mode acceleration method and can be applied to both superelements and the residual structure
when substructuring is employed. Clearly though, there should be sufficient modes comfortably beyond within the
excitation frequency bandwidth to capture the dynamic response, and the static residual vectors should only be used
as a final step to append the quasi-static responses of high frequency (relative to the excitation frequencies) modes.
Note that SOL 111 reverts to a direct solution technique (akin to SOL 108) when there is either viscous or
structural element damping as the equations of motion cannot be orthogonalized. But the unknowns are still the
modal responses and not the physical responses. And because with damping the modes are really complex modes,
there is greater difficulty in capturing all the response using the real modes. Hence, even more modes are thus
required to capture the response accurately in SOL 111 with element damping especially if the elemental damping
is high enough to cause significant complex modes. Thus if elemental damping is high and the modes are complex,
the direct method SOL 108 may be more appropriate.
Another method of determining the adequacy of the finite number of modes is to evaluate the cumulative effective
mass of the chosen number of modes. The effective mass has got a specific meaning in seismic analysis, but can be
used for other dynamic analyses as well as a measure of determining the adequacies of the finite number of modes.
A cumulative effective mass that approaches unity suggests adequacy of the chosen modes to model the spatial
distribution of the loading. This parameter of course does not say much about the adequacy of the chosen modes
to model the frequency content of the loading. Essentially, and to reiterate, the chosen number of modes must be
sufficient to model both the frequency content and also the spatial distribution of the loading. It is found that for
a high cumulative effective mass, uniformly distributed loads such as in seismic analyses require only a few modes
whereas a concentrated patch requires more modes and finally a highly concentrated load requires a much greater
number of modes to capture the effect of the spatial distribution of the loading.
336
Structural Engineering Analysis and Design
Second Edition
4.3.4 Complex Response F(ω) and Amplification D(ω) With Elemental and/or Modal Structural Damping
The response at zero excitation frequency from a forced frequency response analysis (SOL 111 or SOL 108) will
not be equal to the true static response (from a SOL 101 solution) if there is elemental or modal structural damping
within the model. Intuitively, we know this to be true simply due to the fact that structural damping is independent
of forcing frequency. Hence, there is no reason for it to be zero at an excitation frequency of zero. We shall
illustrate this mathematically for modal structural damping although the same concept applies when there exist
elemental structural damping. We know that the (magnitude of the) dynamic amplification factor for a mode when
there is modal viscous damping is
1
D i (ω) =
(1 − ω 2
/ ω ni )
2 2
+ (2ζ i ω / ω ni )
2
If however we have modal structural damping, the modal equation of motion will be
Pi (ω)
ξ i (ω) =
− ω M i + i(C i ω + G iK i ) + K i
2
And hence the (magnitude of the) modal dynamic amplification factor will be
1
D i (ω) =
(
1 − ω 2 / ω ni
2 2
)
+ (G iK i + 2ζ i ω / ω ni )
2
At zero excitation frequency, the (magnitude of the) dynamic amplification is thus not unity, but instead
1
D i (ω) =
1 + (G i K i )
2
337
Structural Engineering Analysis and Design
Second Edition
A SDOF dynamic model is represented by a modal mass M, modal stiffness and modal damping. This SDOF
model can even be modeled as a mass attached to a parallel spring and damper although hand calculations are
sufficient to predict the response of a SDOF system exactly. The natural frequency is a function of the modal mass
and modal stiffness. The modal masses, modal stiffnesses (or natural frequencies) of a MDOF system can be
obtained from a real eigenvalue analysis SOL 103 or a back calculation from a SOL 111 analysis. The modal
damping can be obtained from a complex modal analysis SOL 107.
It will be shown that modes with small values of modal damping can be approximated simply as real modes
(without its imaginary component). Hence, a real modal mass can be used. If however, the mode includes larges
values of modal damping, its complex modal mass must be evaluated and subsequently a complex modal forced
response analysis must ensue.
Estimation of Real Modal Mass From A Forced Frequency Response Analysis Back-Calculation On A
System Without Any Form of Damping
This procedure is shown just to illustrate the dynamic response function. It is certainly as easy to obtain the modal
mass from a real eigenvalue analysis. To determine the modal mass from a forced response back-calculation, the
following procedure is undertaken. We know that
P
F(ω) = D(ω)
K
P P
Hence, the static displacement = = 2
K ω M
In order to compute the modal mass from a back-calculation of the forced response analysis, a SOL 111 should be
performed with EIGRL limiting the calculated modes to only that being considered. Then the relationship above
can be used to estimate the modal mass. Note that the static displacement obtained will be due only to the
eigenvector being considered. The modal mass obtained using this method will correspond exactly to that
calculated from
We know that
P
F(ω) = D(ω)
K
P P
Hence, the static displacement = = 2
K ω M
This shows that the modal mass is a function of the static displacement and not the (magnitude of the) dynamic
amplification at resonance. At resonance, only damping controls the (magnitude of the) dynamic amplification
response D(ω) of the structure. It is however, not true to say that the dynamic response F(ω) is controlled just by
damping as it is also a function of the modal stiffness (or modal mass). At off-resonant frequencies stiffness,
damping and inertia effects become prominent. The static displacement of the structure with or without damping in
reality will be the same as long as elemental damping does not significantly alter the mode shape. At zero
frequency of excitation, damping in reality does not affect the response because there are no dynamic effects.
Hence, from the equation above, it becomes apparent that the modal mass M does not change as long as the natural
frequency of the mode does not change significantly. As a rule, as long as the mode shapes and the natural
338
Structural Engineering Analysis and Design
Second Edition
frequency does not change too much with explicit damping elements (the mode shapes will not change with other
forms of damping, such as modal damping; note that the frequency does not change much unless the modal
damping values are very high), then the modal mass will stay unchanged. This could be an argument to prove that
the modal mass for a mode of a system with low values of damping stays unchanged from the same mode without
damping, which is readily obtainable from a SOL 103 solution.
However, how do we obtain the modal mass from a complex mode if the eigenvector changes considerably from
that without explicit element dampers? A pure real mode has all its terms in phase which each other. Different
modes will of course be out of phase with each other, but every point in a particular mode will vibrate in phase.
And so, for a mode, all points in the structure reach its maximum and minimum at the same phase instant. A
complex mode on the other hand will have different points of the structure reaching its maximum at different phase
instants, i.e. different points of the structure are out of phase with each other for a particular mode. Of course,
different modes are still out of phase with each other. The degree to which these points are out of phase with each
other is a measure of the complexity of the mode. An Argand diagram with all the terms of the complex
eigenvector plotted will exhibit a narrowband if the mode is only slightly complex. The diagram below shows plots
for 4 eigenvector terms. Since there is only a slight phase difference between the terms (angle between the arrows),
the 4 terms define an almost real complex eigenvector.
Im
Re
The following eigenvector plot on the other hand is highly complex because of the large phase angle difference
between the terms of the eigenvector.
Im
Re
The eigenvectors of cable elements for instance can change considerably when explicit dampers with high
coefficients are placed onto them. Then the modal damping values can become very high (such as even 15 to 20%
of critical). If this occurs, the real modal mass is a poor representation of the mode. Instead, a complex modal mass
must be evaluated. Hence, a real modal approach should thus not be used in this case. Instead a (rather
mathematically involved) complex modal forced response analysis is required. Alternatively, a full MDOF analyses
should be undertaken to investigate the response.
SOL 111 reverts to a direct solution technique (akin to SOL 108) when there is either viscous or structural element
damping as the equations of motion cannot be orthogonalized. But the unknowns are still the modal responses and
not the physical responses. And because with damping the modes are really complex modes, there is greater
difficulty in capturing all the response using the real modes. Hence, even more modes are thus required to capture
the response accurately in SOL 111 with element damping. Thus if elemental damping is high, the direct method
SOL 108 may be more appropriate.
339
Structural Engineering Analysis and Design
Second Edition
SOL 111
$ CASE CONTROL SECTION
$ Sets defining grid ids or element ids
SET < Number > = 1 THRU 100, 211, 343, < etc >
$ Grid output of displacement, velocity and acceleration for excitation frequencies
$ SORT1 lists the results by frequency whilst SORT2 lists the results by grid id
DISPLACEMENT(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
VELOCITY(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
ACCELERATION(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
$ Grid output of applied load vector
OLOAD(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
$ Grid output of modal responses (in modal space)
SDISPLACEMENT(<SORT1/SORT2>,<PRINT,PUNCH>,<REAL/PHASE>) = ALL/< Mode Number >
SVELOCITY(<SORT1/SORT2>,<PRINT,PUNCH>,<REAL/PHASE>) = ALL/< Mode Number >
SACCELERATION(<SORT1/SORT2>,<PRINT,PUNCH>,<REAL/PHASE>) = ALL/< Mode Number >
$ Grid output of real eigenvector for the a-set
SVECTOR(<PRINT,PUNCH>) = ALL/<Grid Set ID>
$ Grid output of SPC forces
SPCFORCES(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Grid Set ID>
$ Element output of force, stress and strain
ELFORCE(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Element Set ID>
ELSTRESS(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Element Set ID>
STRAIN(<SORT1/SORT2>,<PRINT,PUNCH,PLOT>,<REAL/PHASE>) = ALL/<Element Set ID>
$ Analysis Cards
SPC = < ID of SPC Cards Defined in Bulk Data >
METHOD = < ID IN EIGRL >
FREQUENCY = < ID OF FREQi >
$ XY plot output
OUTPUT(XYPLOT)
XYPUNCH <DISP/VELO/ACCE> RESPONSE <subcase>/<Grid ID>(<T1/T2/T3><RM/IP>)
XYPUNCH <SDISP/SVELO/SACCE> RESPONSE <subcase>/<Grid ID>(<T1/T2/T3><RM/IP>)
XYPUNCH <ELFORCE/ELSTRESS/STRAIN> RESPONSE <subcase>/<Element ID>(<Code Number>)
$ BULK DATA
Lower Upper Eigenvalue
Number of
EIGRL ID Frequency Frequency Normalization
Eigenvalues
(Hz) (Hz) Method
Number
FREQ1 ID fstart ∆f
of ∆f
flower fupper
FREQ4 ID FSPD NFM
bound bound
All FREQi entries with the same selected ID, selected by the FREQUENCY entry in the Case Control Section, will
be combined. The FREQ1 bulk data entry selects the frequencies at which the frequency response analysis is
performed. It is important to specify a fine enough frequency step size ∆f to adequately predict peak response. Use
at least five to ten points across the half-power bandwidth. For maximum efficiency, an uneven frequency step size
should be used. Smaller frequency spacing should be used in regions near resonant frequencies, and larger
frequency step sizes should be used in regions away from resonant frequencies. FREQ4 defines a frequency spread
FSPD for each and every normal mode that occurs within the bounds of the frequency range with NFM evenly
340
Structural Engineering Analysis and Design
Second Edition
spaced frequencies per spread. Hence, to choose 11 equally spaced frequencies across a frequency band of 0.7fn to
1.3fn for each natural frequency within flower bound and fupper bound, FSPD = 0.30 and NFM = 11. The spread should
definitely be greater than the half-power bandwidth, which for a SDOF system is approximately 2ζfn because ζ ≈
(f2-f1)/2fn hence the spread f2-f1 ≈ 2ζfn.
FREQ3 specifies excitation frequencies in the range between two modal frequencies. The increments between
excitation frequencies are calculated linearly or logarithmically. These frequencies may be clustered towards the
end points of the range of towards the center by specifying a cluster parameter.
Printed .f06 frequency response output can be in SORT1 or SORT2 format. In the SORT1 format, the results are
listed by frequency i.e. for each frequency the results of all grid points are given, whilst in the SORT2 format, the
results are listed by grid point i.e. for each grid point the results of all frequencies are given. In the modal frequency
response analysis, PARAM, CURVPLOT, 1 and PARAM, DDRMM, -1 are required to obtain SORT1 output. To
define frequency frozen structural plots,
DISPLACEMENT(PLOT,PHASE)=ALL
$ BULK DATA
PARAM,DDRMM,-1
PARAM,CURVPLOT,1
For each frequency of excitation, the magnitude of the response at each and every point in the structure is animated
with the phase information. It is meaningless and impossible to plot the real and imaginary components, as the
animation requires the phase difference between different points in the structure. It is prudent to limit the
frequencies of excitation so as to limit the amount of output.
341
Structural Engineering Analysis and Design
Second Edition
Applied load excitations are as described in direct forced frequency response analysis.
4.3.6.1.3 Damping
Viscous and structural modal damping is defined as follows. TYPE refers to the type of damping units, i.e. whether
structural damping G (default), critical damping CRIT or quality factor Q. The values of fi and gi define pairs of
frequencies and damping. Straight-line interpolation is used for modal frequencies between consecutive fi values.
Linear extrapolation is used for modal frequencies outside the entered range. ENDT ends the table input. If
PARAM, KDAMP, 1 then the modal damping is processed as critical damping and if PARAM, KDAMP, -1 then
the modal damping is processed as structural damping, NASTRAN internally making use of ζi = Gi/2 and Qi = 1/Gi
to convert to the required coefficient from that specified by TYPE.
342
Structural Engineering Analysis and Design
Second Edition
4.3.6.2 GL, ML P-∆ (KGA From KEA) Modal Forced Frequency Response Analysis
It is often necessary to incorporate the reduction in bending stiffness of gravity load resisting columns for the
analysis of lateral loads. The following procedure is undertaken.
Phase 1
Perform static analysis (with loads that cause the greatest negative or positive geometric stiffness) based on [KEA]
but include the segyroa.v2001 alter prior to the Case Control Section in order to compute the geometric stiffness
matrix [KGA]1 (and output into a .pch file) based on the generated element loads from the [KEA] static analysis.
Phase 2
A SOL 111 is undertaken based on [KEA] + [KGA]1 with the k2gg=ktjj statement in the Case Control Section and the
outputted .pch file which contains the ktjj matrix in the Bulk Data. The responses at this stage represent the P-∆
(KGA From KEA) response to the dynamic excitation.
The method is valid when only the prestress is judged to affect the geometric stiffness such as in the
compressive preload of building columns due to gravitational loads and the prestressing of extremely taut cables
that sag very little under gravity but not in systems such as suspension bridges. Where lateral loads are large
enough to affect the geometry of the system with prestress, then a repetitive P-∆, SOL 106, implicit dynamic
relaxation SOL 129 or explicit dynamic relaxation must be employed. But in single P-∆ analysis, because cables do
not have much elastic bending stiffness, the initial static preload subcase should only include the prestress and not
gravity as including gravity is the same as solving two linear static problems of stiffness KEA with preload and
gravity as the applied loads respectively. Clearly, in the gravity case, it is nonsensical as the cables do in reality
have differential stiffness (from the prestress) to resist the gravitational force. Prestress in one direction (i.e. along
the axis of cable) will cause a differential stiffness in the orthogonal direction. Gravity acts in the orthogonal
direction and hence cannot be accounted for in the calculation of the prestress in this single P-∆ analysis. To
quantitatively decide if gravity need not be considered in contributing to the differential stiffness of the cables, a
static P-∆ analysis should be carried out, the first subcase being a SOL 101 with only the prestress as applied loads
and the second subcase a P-∆ SOL 101 (i.e. utilizing the induced prestress from the first subcase to form a
geometric stiffness matrix) with both the gravity and prestress included as applied loads. If the difference in the
cable element forces between subcases 1 and 2 is negligible, then gravity has little influence in affecting the
geometric stiffness. If there is a major difference in the cable element force, then clearly, gravity will affect the
geometric stiffness and as such, a repetitive P-∆ , SOL 106, implicit dynamic relaxation or explicit dynamic
relaxation must be used to converge to the true KT. Likewise, in the single P-∆ analysis of multi-storey buildings,
gravity (and only gravity) acts in axis of columns to generate prestress, and the differential stiffness is computed for
the orthogonal direction reducing resistance to lateral wind forces, applied in the second subcase with gravity too.
343
Structural Engineering Analysis and Design
Second Edition
When a static subcase is specified for linear transient response analysis (SOLs 109 and 112) with
STATSUB(PRELOAD), the data recovery is controlled by PARAM, ADSTAT. By default (YES) the static
solution will be superimposed on the dynamic response solution (displacement, stress and SPCForce). The relative
solution can be obtained in reference to the static solution point by PARAM, ADSTAT, NO. No provision is made
for frequency response analysis, because the static responses contribute only to the zero frequency response.
The STATSUB(PRELOAD) computes the differential stiffness due to the prestress and also the follower force. The
follower force is calculated and incorporated by the use of PARAM, FOLLOWK, YES. We know how the prestress
affects the differential stiffness, namely a tensile prestress causing an increase in stiffness. The effect of the follower
force on the stiffness is different. For example, for a cylinder under external pressure critical buckling load may be over-
estimated (even though the mode shapes are similar) in a SOL 105 and the natural frequencies in vibration may be under-
estimated (even though the mode shapes are similar) in a SOL 103 in the absence of follower stiffness. To the contrary,
these observations are reversed in case of centrifugal loads. Centrifugal forces as a constant (static) load are applied by a
Bulk Data RFORCE to any elements that have masses. The follower stiffness due to centrifugal load has the effect of
lowering stiffness (although the centrifugal load tensioning effect increases stiffness), consequently lowering natural
frequencies (even though the mode shapes are similar) in a SOL 103 and lowering the buckling loads (even though the
mode shapes are similar) in a SOL 105. This effect increases as the RPM increases, and it becomes significant when the
RPM is over 1000. For moderately geometric nonlinear analysis, exclusion of follower stiffness affects the rate of
convergence, but the converged solution is correct. For severely geometric nonlinear analysis, it may not be
possible to obtain a converged solution without including follower stiffness. As the geometric nonlinearity
intensifies, so is the effect of follower stiffness. Therefore, inclusion of follower stiffness greatly enhances the
convergence if the deformation involves severe geometric nonlinearity.
344
Structural Engineering Analysis and Design
Second Edition
4.3.6.3 GL, ML P-∆ (KGA From Exact or Approximate KTA) Modal Forced Frequency Response Analysis
It is often necessary to include the differential stiffness, especially if there are prestressed cables in the model. To
obtain KTA, to be theoretically exact, a GNL SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic
relaxation LS-DYNA) with prestress (as temperature loads say) and gravity must be undertaken. Alternatively, an
approximation to KTA can be obtained by repetitive P-∆ static analyses with the prestress (as temperature loads say)
and gravity applied. The procedure to obtain this approximate KTA will be presented. Note that the approximate KTA
will be the summation of the elastic stiffness KE at the undeflected (by the prestress and gravity) state but KG at the
deflected (by the prestress and gravity) state. Hence if KE changes considerably during the application of the
prestress, a full SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-DYNA),
which converges to the KE and KG at the deflected (by the prestress and gravity) state should be employed. Hence
for the modelling of a suspension bridge where there is a great change in geometry (known in the bridge industry as
form-finding, so-called because it is necessary to find the form or shape of the catenary suspension cables), it may
be prudent to employ SOL 106 (or implicit dynamic relaxation SOL 129 or explicit dynamic relaxation LS-
DYNA), but for a high tension low sag cable on say a tower with prestressed cables, the repetitive P-∆ static
analysis may be adequate. The repetitive P-∆ analysis basically involves a number of iterations of linear static
analyses to obtain an approximate KTA. Note again that A refers to the initial undeflected (by the collapsing load)
state, but deflected by the prestress and gravity. To perform the repetitive P-∆ analysis, a static analysis is
performed based on KEA with temperature loads and gravity to generate forces in the structural elements, which in
turn provides input for the computation of KGiAKTm where m is the iterations. Repetitive static analysis is performed
with the prestress and gravity updating the stiffness matrix KEA + KGiAKTm-1 + KGiAKTm until convergence of
displacements is obtained. The tangent stiffness at this stage is the approximate converged tangent stiffness matrix
KTA = KEA + KGiAKT. The converged displacements represent the approximate P-∆ (KGA From Approximate KTA)
static response to the initial prestress loads. The converged geometric stiffness at this stage would be that based
upon the approximate tangent stiffness matrix KTA, i.e. KGiAKT.
Phase 1
Perform static analysis (with prestress and gravity) based on [KEA] but include the segyroa.v2001 alter prior to the
Case Control Section in order to compute the geometric stiffness matrix [KGA]1 (and output into a .pch file) based
on the generated element loads from the [KEA] static analysis.
Phase 2
Perform static analysis (with prestress and gravity) based on [KEA] + [KGA]1 by including the k2gg = ktjj statement
in the Case Control Section, the outputted .pch file which contains the ktjj matrix in the Bulk Data and the
segyroa.v2001 alter prior to the Case Control Section to compute the [KGA]2 (and output into the .pch file
overwriting previous data) based on the generated element loads from the [KEA] + [KGA]1 static analysis.
Phase 3
Repeatedly perform the Phase 2 static analysis (with prestress and gravity) based on [KEA] + [KGA]i for i = 2 to n
where n represents the number of iterations required for the change in deflections between analyses to become
negligible. This would signify that the change in the [KGA] matrix become negligible and the correct [KGA] is
attained. The deflections and the other responses at this stage represent the P-∆ (KGA From Approximate KTA) static
response to the prestress and gravity. The stiffness of the structure is KTA.
Phase 4
A SOL 111 is undertaken with the k2gg=ktjj statement in the Case Control Section and the outputted .pch file
which contains the latest ktjj matrix in the Bulk Data. The responses at this stage represent the P-∆ (KGA From
Approximate KTA) response to the dynamic excitation.
345
Structural Engineering Analysis and Design
Second Edition
4.3.7.1 Determination of Maximum Dynamic Displacement for Deterministic Frequency Domain Loading
by Transforming the Coupled MDOF Linear Damped ODEs To a Set of Independent (Uncoupled)
SDOF ODEs and Solving the Independent Equations in a Manner Similar to Solving a SDOF ODE
A coupled MDOF system of linear ODEs must be uncoupled to a set of independent SDOF system of linear ODEs
if any feasible hand method computation is to be employed. These independent SDOF ODEs can be solved using
standard techniques of solving SDOF linear dynamic ODEs. The final stage to the analysis is to employ a modal
superposition method to express the total response as a summation of the results from the solution of the individual
modal equations. Note that since the modal frequencies and the corresponding mode shapes are required to
uncouple the coupled MDOF ODEs, the real eigenvalue analysis must be performed first. It was shown that a
system of coupled MDOF ODEs could be reduced to a system of independent SDOF ODEs by employing the
orthogonality properties of real modes
C i = {φ i }T [C]{φ i } = ζ i 2M i ω n i
K i = {φ i } [K ]{φ i } = ω 2ni M i
T
Pi (ω) = {φ i } {P(ω)}
T
The modal properties (i.e. modal masses, modal frequencies and hence modal stiffness) can be obtained by
performing a real modal analysis SOL 103 and if required the modal damping can also be incorporated from a
complex modal analysis SOL 107. The normalization of the mode shapes can be arbitrary. The normalization
technique employed will not affect the value of the modal frequencies but of course will determine the values of the
modal masses (and hence modal damping and stiffnesses) and the (amplitude of the) modal force. Although the
normalization technique is arbitrary, it is recommended that the normalization employed would facilitate the
calculation of the (amplitude of the) modal force since this is a hand calculation. To facilitate the computation, it is
wise to choose the normalization to be unity at the DOF of application of the external excitation. Either way, the
modal force (for a particular mode) for discrete loading points is calculated as follows
Pi (ω) = {φ i } {P(ω)}
T
Had the loading been continuous, a continuous modal force can also be calculated by hand (or a spreadsheet) as
follows
L
Pi (ω) = ∫ φ i ( x ){P( x , ω)}dx
0
For the special case, which is most common in reality, when the complex harmonic force vector is independent of
the frequency of excitation and simply real, we have {P(ω)} = {p0} and hence,
[ ]
{P( t )} = Re al {P(ω)}e iωt
= Re al [{p }e ]
0
iωt
hence Pi (ω) = {φ i } {p 0 }
T
= p 0i
346
Structural Engineering Analysis and Design
Second Edition
There will be n uncoupled ODEs for n natural modes. These are now simply SDOF ODEs, which can be solved
using complex arithmetic. Hence, the complex modal response (in modal space)
Pi (ω) p 0i
ξ i (ω) = =
− ω M i + iC i ω + K i − ω M i + i C i ω + K i
2 2
p 0i − iθi 2ζ i ω / ω ni
ξ i ( ω) = D i ( ω) e , θ i = tan −1
Ki 1 − ω 2 / ω ni(2
)
where the real modal amplificat ion factor and its maximum are
1 1
D i (ω) = ; D i max =
(
1 − ω 2 / ω ni
2 2
+ (2ζ i ω / ω ni ) )
2 2ζ 1 − ζ 2 ( )
Thus, the steady-state solution in the time domain
p 01
D1 (ω) cos(ωt - θ1 )
K1
...
p 0i
{u (t )} = [{φ}1 ... {φ}i ... {φ}n ] D i (ω) cos(ωt - θi )
Ki
...
p 0n
D n (ω) cos(ωt - θ n )
K n
347
Structural Engineering Analysis and Design
Second Edition
Note that p0i corresponds to the amplitude of the modal loading function Pi(ω) and not simply the amplitude of the
loading function {P(ω)}, i.e. p0 at any particular node. Of course, if the normalization of the mode was such that it
was unity at the DOF of application of the external force with amplitude p0 then, p0i will be equal to p0.
p0
D1 (ω) 1 cos(ωt - θ1 )
u 1 ( t ) φ11 φ12 K1
=
u 2 ( t ) φ 21 φ 22 p
D 2 (ω) 2 cos(ωt - θ 2 )
0
K2
The above computation takes into account the phase difference between the responses of different modes. It is
important to point out that the maximum of the modal responses ξi(ω) do not occur at the same time. The frequency
domain response ODE can only be solved for a particular excitation frequency ω at a time. For a particular ω there
will be different responses ξi(ω) from different modes, differing in amplitude of response Di(ω), static displacement
and in phase θi. Because there is a difference in phase angle between the responses of each and every mode from
2ζ i ω / ωni
θ i = tan −1
(1 − ω 2
/ ωni
2
)
the exact solution involves maximizing a complicated trigonometric expression with many phase angles. The exact
solution can be obtained by expanding the general u(t) expression to include all the modes considered, and only
then maximised.
However, this may prove to be mathematically involved if differentiation is to be performed analytically by hand
(the computerized method of course performs the response calculation at discrete frequency points as it is not just
the maximum that is of interest). An alternative would be to instead of maximising the physical response, maximise
the modal responses and then only superpose the modal responses for the physical response.
1
D i max =
(
2ζ 1 − ζ 2 )
p 0i
ξ i max = D i max (ω)
Ki
This is equivalent to the response spectrum method (only that the procedure is employed on a deterministic loading
function instead of a random function). The square root of sum of squares (SRSS) of the maximum modal
contributions as follows, depicted for a 2 DOF dynamic system.
Alternatively, the Complete Quadratic Combination (CQC) method is used if the modal natural frequencies are too
close to each other (within about 10 %). This is based on random vibration theory. Of course, the upper limit
combination would be to choose all the eigenvectors to have the maximum modal response at the same time.
Often, we need to ascertain other types of response such as velocity and acceleration.
We know that,
p 0 i ( ωt -θ)
{u ( t )} = Re al [Φ ]D(ω) e
K
348
Structural Engineering Analysis and Design
Second Edition
The differentiation of the complex vector representing the displacement is equivalent to multiplying the length of
the vector by ω and turning it ahead by 90 degrees. Multiplying any complex number by imaginary i is equivalent
to a rotation of 90 degrees. Hence the velocity leads the displacement by 90 degrees and the acceleration leads the
velocity by 90 degrees. The magnitude of the velocity is ω times that of the displacement and the magnitude of the
acceleration is ω times that of the velocity. In all cases, it is implicitly understood that it is the real part of the
complex number that represents the physical harmonic motion.
Note that the total response is obtained by multiplying the eigenvectors by their corresponding modal amplitude
factor, which is an indication of the prominence of the mode in the response. It is often the case that lower modes
of vibration will have greater prominence. Hence, only a few of the initial prominent modes need to be evaluated in
the expression. However, an important point to observe is that since the velocity and acceleration total responses in
the time domain are a function of the frequencies and the square of the frequencies respectively on top of the usual
modal amplitude factors, this would mean that the prominence of higher modes become more significant when
velocities and accelerations are computed.
Ideally, the responses in the time domain should be maximised to obtain the maximum response. But if this proves
to be difficult, the modal responses can be maximised and combined using some superposition method such as
SRSS or CQC (effectively a response spectrum analysis on a deterministic loading function). For whatever
response be it the displacement, velocity acceleration or force, the individual modal responses should first be
calculated in terms of time t, then maximised, then only combined for the physical response with some method of
superposition. Do not maximise the modal displacement response and base the other velocity and acceleration
responses on that.
349
Structural Engineering Analysis and Design
Second Edition
A forced response analysis of a shear frame (be it to model a building or water tank) is illustrated in MAPLE 7. All
units in SI.
M
u1 10000cos4.5826t
EI = 560000000
M = 400000
2L EI L=4
M M
u2
EI L
2L EI
350
Structural Engineering Analysis and Design
Second Edition
5.728219619 0.
wn :=
0. 11.45643924
2 -1
Φ :=
1 1
1 -1
Φ := 1
1
2
Uncoupling The Equations of Motion
> MODALMASS:=evalf(Transpose(Phi).MASS.Phi);
MODALSTIFFNESS:=evalf(Transpose(Phi).KE.Phi);
MODALFORCE:=evalf(Transpose(Phi).P(w));
600000. 0.
MODALMASS :=
0. .1200000 10 7
8
.19687500 10 0.
MODALSTIFFNESS :=
0. .157500000 109
10000.
MODALFORCE :=
-10000.
Modal Responses Re[ξi] = Di(w)Poi/Ki cos(wt-θi)
> D(w):=Vector(DOF):
for i from 1 to DOF do
D(w)[i]:=1/((1-w^2/lambda[i,i])^2+(2*MODALDAMPING[i,i]*w/lambda[i,i]^(1/2))^2)^(1/2);
end do;
xi:=Vector(DOF):
for i from 1 to DOF do
xi[i]:=D(w)[i]*MODALFORCE[i,1]/MODALSTIFFNESS[i,i]*cos(w*t-
arctan(2*MODALDAMPING[i,i]*w/lambda[i,i]^(1/2)/(1-w^2/lambda[i,i])));
end do:
xi:=evalf(xi);
D( 4.5826 ) := 2.775090553
1
D( 4.5826 ) := 1.190424609
2
351
Structural Engineering Analysis and Design
Second Edition
An absolutely equivalent method of solving the equations in the frequency domain is to work in real and imaginary
terms instead of in terms of magnitude and phase.
p 0i
=
− ω M i + K i + iC i ω
2
p 0i / K i
=
(1 − ω 2
)
/ ω ni + i(2ζ i ω / ω ni )
2
Instead of the polar form, this can be written in terms of the real and imaginary parts
ξ i (ω) =
p 0i
(
1 − ω 2 / ω ni
2
−i
)
(2ζ i ω / ω ni )
(
K i 1 − ω 2 / ω 2 + (2ζ ω / ω )2
ni
2
i ni )1 − ω 2
/ ω ni
2 2
+ (2 ζ i ω(/ ω ni )2
)
ξ i (ω) = ξ